Abstract
eBay’s ID Verify program, which allowed sellers to pay a fee to have their identity confirmed by a credit information company, was designed to alleviate asymmetric information problems in online auctions. We conduct a field experiment where items are sold on eBay by one of four identities that either are or are not ID verified and have either a good or no reputation. We find that ID verification increases the number of bids placed but has no impact on the level of the winning bid, and may even lower the winning bid among sellers who have good reputations.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
For example, the White House instituted an initiative in April 2011 called the National Strategy for Trusted Identities in Cyberspace (NSTIC), the goal of which was “to protect individuals, businesses, and public agencies from the high costs of cyber crimes like identity theft and fraud, while simultaneously helping to ensure that the Internet continues to support innovation and a thriving marketplace of products and ideas.” In the private sector, IDology Inc., for example, offers a service called Expect ID, which “instantly validates an identity to ensure transactions move forward quicker and without manual intervention.” More information can be found at http://www.idology.com/id-verification/id-verification.
Although the service has been discontinued, users who were ID verified still have the icon displayed next to their username.
eBay’s reputation system has been thoroughly studied. Lucking-Reiley et al. (2007) offered the first study of the effect of eBay's reputation system on auction outcomes. While this paper was not published until 2007, the initial working paper was made available in 1999 and remains the most widely cited study of reputation on eBay. Many others have followed; Bajari and Hortaçsu (2004) provide a survey of the results and Hasker and Sickles (2010) offer a more recent survey of the variety of issues that have been studied using data from eBay transactions.
For example, in the three samples of eBay auctions that were examined by Livingston (2010a) that were collected after the ID Verify service was introduced, there is not a single observation where the seller had a feedback rating of 0 and was ID verified.
As of April 6, 2018, the press release could still be viewed here: http://www.prnewswire.com/news-releases/ebaytm-selects-equifax-secure-to-verify-identity-of-buyers-sellers-over-the-internet-73424207.html.
Private communication with eBay.
The good-reputation identities ID0Rep1 and ID1Rep1 began the experiment with the exact same feedback rating of 18 (18 positive reviews, and zero negative reviews) which was developed through our natural personal participation in the eBay marketplace. Ratings of this magnitude are found to have a substantial, positive effect on auction outcomes relative to a feedback rating of zero by Livingston (2005) and Livingston (2010b).
The feedback profiles of the good-reputation identities are similar on many dimensions other than just rating: Both have no negative reports; and both generated their feedback almost entirely through buying rather than selling. While ID1Rep1 was registered approximately 1 year earlier and generated a portion of its feedback earlier, neither identity had received any ratings in the 6 months prior to the beginning of the experiment.
The entire experiment lasted longer than 30 days because ID1Rep1—the ID verified seller identity with a solid feedback rating—was (ironically) suspended by eBay from starting new auctions from July 4 through July 12, because the sudden increase in recent sales from a new computer was viewed as suspicious. The experiment recommenced on July 13, when the identity was reinstated. However, it had no effect on the qualitative conclusions of the analysis. Accordingly, in the analysis that follows the results that use the entire sample are reported, and this event and break in the experiment are not controlled for.
Roth and Ockenfels (2002) show that because eBay auctions end at a pre-specified time, much of the bidding activity occurs in the closing moments of an auction. Spacing the end times of the auctions 2 h apart thus promotes minimal overlap in the most intense periods of bidding in each auction in the sample while maintaining common market conditions across the auctions as much as possible. This strategy was fairly successful: 90% of the auctions in the sample had their winning bids placed during the last 2 h of the auction; and 75.2% of bids that were within $5 of the winning bid were placed in the last 2 h of the auction in which they were submitted. However, since the auctions lasted for 3 days, there was some overlap of the bidder pools. 21.5% of the unique bidders in the sample submitted bids to more than one auction.
Examples of the appearance of each page are available from the authors by request.
Five buyers won two auctions that were part of the experiment. Of these five, two bought from the same seller ID, and three bought from different seller IDs. The former are a particular concern because they were likely aware that the auction was part of the experiment when placing their bids in the second auction that they won. This possibility is accounted for in the analysis that follows.
But again, as Livingston (2005) shows, a small number of positive ratings has a large effect on bid amounts, but further reports have little marginal impact. We therefore expect these additional reports beyond 18 to have little if any impact on auction prices.
All analysis is conducted excluding an outlier where two bidders entered a bidding war and the winner paid a price of $77.00, which is $27.01 over the retail price of the good and $24.11 higher than the next highest winning bid of $52.89. All analysis thus uses 119 observations. Including this observation has no impact on the qualitative results.
One potential reason might be that the bidder advantage does not persist when attention is restricted to “serious” bidders. Frequently in eBay auctions, some bidders test the waters by offering extremely low bids early in the auction. Such bidders may have done so more frequently in the auctions of the good-reputation ID verified seller. For example, when attention is restricted to only bidders whose final bids were no more than $5 below the lowest selling price in the sample ($24.50), the number of bidders participating in the auctions of each identity are not statistically different, and the mean number of bidders are very similar (ranging from 4.97 to 5.23). This analysis is omitted from the paper for brevity, but is available from the authors by request.
The order in which the auctions of each identity ended was randomized by date, so identity characteristics were effectively randomized within each date. Date fixed effects ensure that the estimated effects are calculated using within-date variation, but standard errors are also clustered by date since Cameron and Miller (2015) note that fixed effects may not fully control for within-cluster correlation. They also note that when cluster sizes are small—as they are here since four auctions end on each date—fixed effects should be controlled for using the within-regression estimator: the mean-differenced model. We follow this recommendation for all specifications that control for date fixed effects.
The results are qualitatively similar if the equation is estimated instead by OLS. This is important to note since one assumption of the Poisson model may not be met: The participation of each bidder should be independent, but it is possible that the activity of previous bidders could attract or discourage other bidders from participating.
It is not possible to conduct an F test of the joint significance of the date fixed effects because the number of regressors that vary within cluster is less than the number of clusters minus one, as is the case here. See Cameron and Miller (2015).
We thank Lawrence White for suggesting this line of inquiry.
To conduct the LR test, we add interaction terms between REPit and TWOBIDSit, the date fixed effects, and the order fixed effects to the equation for the pooled sample, Eq. 5, to allow all coefficients except that on IDit to vary across the subsamples.
This test is performed by running an OLS regression of the winning bid on IDit, REPit, and their interaction, the number of bidders, and date and order fixed effects, and performing a joint test of significance of the effects of REPit and its interaction with IDit.
This test is based on the same regression as the previous test, and is conducted by performing a joint test of significance of the effects of IDit and its interaction with REPit.
Most of the extant empirical analyses of eBay auctions (which follow the lead of Lucking-Reiley et al. 2007) typically omit the number of bidders as a control from price regressions and otherwise ignores the potential endogeneity. While it may be the case that the number of bidders is exogenous in the sample of auctions we conducted as part of our experiment, it is likely endogenous in samples of the type that are typically studied in the literature. These samples typically include all auctions of a given product within a certain timeframe and many that end at similar times, which makes it more likely that these auctions compete with each other for bidders. We are hopeful that future studies can employ the methodology we introduce here to account for this endogeneity.
In the sample, the correlation coefficient between TWOBIDS and the number of participating bidders is -0.486, and the correlation coefficient between TWOBIDS and the winning bid is − 0.017.
The files that contained the bid information for the other six auctions were not saved properly, and this was not noticed until after the bid information files were removed from eBay’s servers.
The results are qualitatively similar if all bids are considered.
Note that a bidder can place a bid in more than one auction in the block.
A post that discussed ID Verfiy (which is no longer available) had been viewed 19,000 times; but at the time of the post in November 2007, it stated that “Most people have not heard of the eBay ID Verify Program.”
We thank an anonymous referee for offering this explanation.
References
Anwar, S., McMillan, R., & Zheng, Mingli. (2006). Bidding behavior in competing auctions: Evidence from eBay. European Economic Review, 50(2), 307–322.
Bajari, P., & Hortaçsu, A. (2004). Economic insights from internet auctions. Journal of Economic Literature, 42(2), 457–486.
Cameron, A. C., & Miller, D. L. (2015). A practitioner’s guide to cluster-robust inference. Journal of Human Resources, 50(2), 317–372.
Harrison, G. W., & List, J. A. (2004). Field experiments. Journal of Economic Literature, 42(4), 1009–1055.
Hasker, K., & Sickles, R. (2010). eBay in the economic literature: Analysis of an auction marketplace. Review of Industrial Organization, 37, 3–42.
Livingston, J. A. (2005). How valuable is a good reputation? A sample selection model of internet auctions. Review of Economics and Statistics, 87(3), 453–465.
Livingston, J. A. (2010a). Functional forms in studies of reputation in internet auctions. Electronic Commerce Research, 10(2), 167–190.
Livingston, J. A. (2010b). The behavior of inexperienced bidders in internet auctions. Economic Inquiry, 48(2), 237–253.
Lucking-Reiley, D., Bryan, D., Prasad, N., & Reeves, D. (2007). Pennies from eBay: The determinants of price in online auctions. Journal of Industrial Economics, 55(2), 223–233.
Peters, M., & Severinov, S. (2006). Internet auctions with many traders. Journal of Economic Theory, 130(1), 220–245.
Roth, A. E., & Ockenfels, A. (2002). Last minute bidding and the rules for ending second-price auctions: Theory and evidence from a natural experiment on the internet. American Economic Review, 92(4), 1093–1103.
Acknowledgements
We thank Lawrence White, Dhaval Dave, and two anonymous referees—all of whom provided many helpful comments and suggestions. Brian Johns, Bidisha Ghosh, and Yunlei Tu provided excellent research assistance. Funding was provided by the Bentley University Faculty Affairs Committee. Of course, all remaining errors are our own.
Author information
Authors and Affiliations
Corresponding author
Appendix: Bidder Communication
Appendix: Bidder Communication
1.1 Invoice Email
Congratulations on winning my auction for an iPod shuffle! Click the “Pay Now” button above for payment details. As soon as payment is received, it will be shipped. Payment by Paypal will speed up shipping considerably.
This auction was part of an economics experiment. I ask that you please do NOT leave me any feedback! I need to keep my feedback rating unchanged during the experiment for it to work. Thanks!
1.2 Shipping Notification Email
Hi (name here),
Your iPod Shuffle was shipped this morning to:
(address here)
Thanks for your business!
This auction was conducted as part of an economics experiment, so I ask that you NOT leave me any feedback! For the purposes of the experiment, I need to keep the feedback rating of the ID used to sell the iPod unchanged. Thanks much!
1.3 Shipping Insert
Hello,
Enclosed is your new iPod Shuffle. Thank you for your purchase!
This auction was run as part of an economics experiment. I ask that you please do NOT leave me feedback! I need to keep the feedback rating unchanged throughout the experiment for it to work. Thanks!
Rights and permissions
About this article
Cite this article
Livingston, J.A., Scholten, P.A. The Effect of ID Verification in Online Markets: Evidence from a Field Experiment. Rev Ind Organ 54, 595–615 (2019). https://doi.org/10.1007/s11151-018-9655-7
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11151-018-9655-7