At a time in which online platforms are being utilized in any number of different sectors, this particular study was based on a field experiment which most of us don't have the opportunity to conduct.
Creating Low-Code Loan Applications using the Trisotech Mortgage Feature Set
A field experiment: Growing Two-Sided Networks By Advertising The User Base
1. - 1 - Laurence (Larry) J. Pino, (Esq.)
Growing Two-Sided Networks by Advertising the User
Base: A Field Experiment
- 1 - Laurence (Larry) J. Pino, (Esq.)
2. - 2 - Laurence (Larry) J. Pino, (Esq.)
The Authors
Catherine Tucker
Professor of Marketing
MIT Sloan School of Management
PhD, Economics
Stanford University
Juanjuan Zhang
Professor of Marketing
MIT Sloan School of Management
PhD, Business Administration
University of California, Berkeley
3. - 3 - Laurence (Larry) J. Pino, (Esq.)
Purpose of the Study
A two-sided exchange network is a web-based platform which provides transactional
capability between two parties, typically a buyer and a seller. Commonly known two-
sided exchange networks are companies like OpenTable.com, VRBO.com, Monster.com,
and others. When set up properly, the two-sided exchange networks like Ebay.com and
Match.com experience explosive growth. On the other hand, if not structured
effectively, such as Chemdex.com and others, the companies flounder and ultimately
shut down. Critical to the success of a two-sided exchange network is growing network
participation by properly advertising the size of their user base. As critical as the
various informational display formats reflecting the number of buyers, sellers and
transactions might be, according to the authors, no research is extant guiding how two-
sided networks should advertise that information relying, instead, on
website designers, interns and graphic artists viewing the
display as a GUI function, as opposed to viewing the
decision as a strategic marketing variable. Note Table 1
following for current formats. As such, the purpose of the
Study is to provide useful data informing firms on how
they can effectively use “strategic information revelation as
a marketing tool to build two-sided networks.”
4. - 4 - Laurence (Larry) J. Pino, (Esq.)
Purpose of the Study
5. - 5 - Laurence (Larry) J. Pino, (Esq.)
Research Landscape of the Study
While no research is available in the literature to address the issue of whether two-
sided networks should be advertising the number of buyers, the number of sellers, the
number of transactions, or some combination of all of them, there is research
addressing additional components of the general field of information disclosure.
The authors have defined the information display formats as an “indirect network
externality” and, in so doing, rely on literature associated with two-sided networks
(Bucklin and Sismeiro, 2003; Ellison and Ellison, 2005)
In a related area, the authors argue that their findings are also related to the “retail
colocation” literature. For example Dudey (1990) and Wernerfelt (1994) “show that
competing retailers may choose to concentrate in the same shopping mall as a
commitment not to raise prices to exploit consumers’ sunk cost of traveling.”
6. - 6 - Laurence (Larry) J. Pino, (Esq.)
Research Landscape of the Study
(con’t) Additionally, Iyer and Pazgal (2003) “demonstrate that the gathering of competing
sellers through ‘Internet shopping agents’ can mitigate price competition as a larger
number of competitors decreases each seller’s chance of winning price-sensitive
shoppers.
• In other areas, the literature has addressed factors that affect network participation:
Fath and Sarvary (2003), for example, identify one particular growth strategy for B2B
exchanges, which is to subsidize buyers. And, Chen and Xie (2007) discover that the
“lack of customer loyalty can end up benefitting firms in the presence of cross-
market network effects.”
In short, the authors contribute to the research landscape by focusing in general on
information display formats and in particular on information associated with the
number of buyers, the number of sellers, and the number of transactions, for the
purpose of ascertaining what is the optimal marketing strategy for the purpose of
effectively growing a two-sided exchange platform.
7. - 7 - Laurence (Larry) J. Pino, (Esq.)
Data and Methodology
Field Experiment
The field experiment focused on a B2B website that resembles Craigslist.org, although
its actual name and location are protected by a Confidentiality Agreement. The
website receives 240,000 clicks per day and provides a common platform for sellers of
various types of goods and real estate properties listed in multiple specific categories.
The field experiment ran from November 29, 2006 to January 15, 2007 in the largest
city market that the website served out of the forty major metropolitan areas the
website served. A total of 3,314 attempted listings across fifteen categories were
exposed to the experiment.
In order to post a listing, a seller must register a user account, login to that user
account, and fill out a listing form. While there is no actual fee charged for a listing,
there is time spent which sellers must compare to the expected return (Fath and Sarvary,
2003). As a result, the seller attrition rate between the time a listing is started and the
submission is 16%.
8. - 8 - Laurence (Larry) J. Pino, (Esq.)
Data and Methodology
Nature of Treatments
The website randomized display of the number of sellers and/or buyers to reach
potential sellers and it randomized how many sellers/buyers it actually stated. The text
content displayed with respect to those randomized displays was drawn from the
following four treatment conditions:
1) “Presently, there are [#Sellers] listings and [#Buyers] users viewing these
listings in the [category name] category of [city name].”
2) “Presently, there are [#Sellers] listings in the [category name] category of [city
name].”
3) “Presently, there are [#Buyers] users viewing these listings in the [category
name] category of [city name].”
4) (A blank page.)
9. - 9 - Laurence (Larry) J. Pino, (Esq.)
Data and Methodology
Statistical Checks
The authors ran a number of statistical checks as follows:
• The #Sellers and #Buyers were randomized for each potential seller to protect
against confounds from unobservable variables.
• #Sellers and #Buyers were drawn from a uniform distribution between 1 and 200.
• The ambiguous word “presently” was used to allow some vagueness in order to
avoid deceiving customers through the randomization procedure.
• Regressions were used to confirm that the randomization procedures were
implemented correctly.
• The authors clustered standard errors at the category level in model
estimations which included category, week, day, hour, and number
of visits in order to control for any departure from full randomness.
• The authors retained data on each seller’s first visit on each day, but
removed data for subsequent visits on the same day in order to
parallel as best as can be done in a website field experiment a pure
between-subjects design.
10. - 10 - Laurence (Larry) J. Pino, (Esq.)
Statistical Checks (con’t)
• In the matching process, 128 observations out of a net of 3,314 observations were
unable to be matched and were therefore excluded from the empirical analysis.
• Spammers, defined as a seller who submitted more than ten listings in the same
category during the experimental period, were eliminated which resulted in the
removal of 1,509 listings. To the extent that a spammer or a bot posts the listing
irrespective of what information is displayed, an action taken by a spammer or bot
was not relevant to the purpose of this Study.
Data and Methodology
• And finally, the authors ran a seemingly
endless and unpronounceable series of
additional statistical checks apparently to
do something incomprehensible and
presumably academically career-enhancing
to produce statistical results totally
indecipherable.
11. - 11 - Laurence (Larry) J. Pino, (Esq.)
Data and Methodology
Data (Informational Display Formats)
The authors collected two data sets which were used for the analysis: A click-stream
data set and a treatment data set.
• A click-stream data set consisted of a time stamp, the users IP address, a record of
all Web page requests, and an error code.
• The treatment data set contained an IP address, a time stamp, the product category
the potential seller intended to list in, whether information on the number of
buyers and/or sellers was displayed, and the actual number of buyers and/or sellers
drawn if applicable.
• The click-stream data set was compared to the treatment data set using the IP
address and the time stamp and, based on that comparison comparative
correlations were evaluated.
12. - 12 - Laurence (Larry) J. Pino, (Esq.)
Data and Methodology
Data (Informational Display Formats)
13. - 13 - Laurence (Larry) J. Pino, (Esq.)
Data and Methodology
Data (Aggregate Economic Effects of Different Display Formats)
While the primary purpose of the Study was to provide data to assist two-sided
networks to identify what informational display formats were the most effective at
growing traffic, the authors also attempted to append to the Study an evaluation of the
economic effects of the differences obtained in their correlation data which could be
reduced to total cost savings in the acquisition cost per new seller.
• For that purpose, the authors relied on the actual traffic of the website replacing
#Sellers and #Buyers with the average actual number of sellers and buyers over the
two weeks prior to the experiment in order to approximate the word “presently.”
• They did so acknowledging the difficulties with that estimation and replacement
process.
Buyers Sellers
14. - 14 - Laurence (Larry) J. Pino, (Esq.)
Data and Methodology
Data (Aggregate Economic Effects of Different Display Formats)
• Recognizing that this particular field experiment dealt with categories in just one
regional market chosen for the experiment accounting for just 16% of the total
traffic, the authors extrapolated their underlying statistical results to what would
have happened had the findings been extended to the entire website for all
markets.
• That data multiplied the fitted listing probability for a particular category with
management’s forecast of potential new sellers in that category over a 12-month
period of time.
• Finally, the authors utilized a proxy for seller acquisition costs by utilizing the cost of
search advertising for a particular category in the entire country where the website
is based.
• Thereafter, in an attempt to generalize the implications (even though the
projections are based on a bootstrapping methodology), they conducted 1,000
replications of a randomized draw from the data with replacement to obtain
standard errors for the cost savings estimates.
• The results are identified in Table 4, which follows.
15. - 15 - Laurence (Larry) J. Pino, (Esq.)
Data and Methodology
16. - 16 - Laurence (Larry) J. Pino, (Esq.)
Summary of Data
17. - 17 - Laurence (Larry) J. Pino, (Esq.)
Results
Based upon the Study, the authors concluded as follows:
• Irrespective of the treatment condition, the probability a seller will list is 85% where
information is displayed and 84% where no information is shown.
• When information about both buyers and sellers is displayed, the existence of many
sellers deters further seller listings.
• However, this deterrence effect disappears when only the number of sellers is presented.
• A large number of buyers is more likely to attract new listings when it is displayed together
with the number of sellers.
• Displaying either seller or buyer information in isolation is
more effective than displaying both concurrently.
• The number of new sellers acquired is positively and
significantly correlated with buyer browsing time within a
category when only seller-side information is displayed.
• Providing more information does not always translate into
attracting more customers.
• The efficacy of the display formats varies by category.
18. - 18 - Laurence (Larry) J. Pino, (Esq.)
Results
• Seller information in isolation achieved Total Cost Savings of $27,992.
• Buyer information in isolation achieved Total Cost Savings of $32,992
• Displaying both seller and buyer information resulted in a loss of $85,096 in Total
Customer Acquisition Costs.
19. - 19 - Laurence (Larry) J. Pino, (Esq.)
Implications & Conclusions
The results suggest that two-sided networks’ optimal way of advertising the user base
should take into account potential customers’ knowledge about the network and buyers’
need for comparison shopping.
1. The authors have provided some very substantial and concrete results emanating
from a relatively tortuous methodology for identifying informational display formats.
That information is concrete and actionable. While the data themselves are not
substantially compelling, they are certainly directional and provide a basis for
rational managerial decision-making. That, in itself, is relatively supportable.
2. The Authors attempt to extrapolate those data results into a reduction of acquisition
costs per new seller appears somewhat tenuous to the extent that the analysis is
based primarily on generalized extrapolations, substantive managerial evaluations,
and a proxy which is, in itself, tautological. It would be
sufficient to conclude, based on the data, that there does
tend to be an optimal configuration of the informational
display format for the number of sellers, the number of
buyers, and the number of transactions, without
attempting a more rigorous approach to monetizing what
the cost savings are for the focal company.
20. - 20 - Laurence (Larry) J. Pino, (Esq.)
Implications & Conclusions
3. While the conclusions from the data also attempt to “shed light on some
documented ambiguities surrounding the effect of competition on entry,” the data
themselves do not necessarily have a great deal to do with any of that field of
research. Therefore, references to Camerer and Lovallo (1999); Toivanen and Waterson
(2005); Simonsohn (2010); Narasimhan and Zhang (2000); Debruyne and Reibstein (2005); Camerer
et al (2004); Wernerfelt (1995) might be appropriate as academic references, but shed
relatively little light on either the theoretical construct for the Study or the value of
the takeaways to academic scholars or managerial practitioners. In fact, they could
be viewed as a real head-scratcher.