Our four treatments vary in their network density as follows:
• In NO, the baseline treatment, neither buyers nor sellers have any information about
the past. This is the only treatment without identification of sellers making interaction
completely anonymous.
• Introducing labeling of sellers, in DEGENERATE all buyers know from whom they
shopped in the past. Thus, sellers now have to start thinking about direct reputation
they establish with buyers.
• Increasing network density, in PARTIAL all buyers not only know their own
experience but, additionally, also know the experience of one other buyer in the
market. Thus, sellers may be concerned with indirect reputation effects which are
mediated through consumers’ network relation.
• Finally, in FULL we make the entire history of all sellers available to everybody. As
already shown in Bohnet et al. (2005) this is the treatment that achieves the best
market outcomes that the provision of feedback about sellers’ past can achieve on its
own.2
In each treatment, subjects play this game for 30 periods. Each matching group consists of
four buyers and four sellers who keep their roles throughout the experiment. At the beginning
of each period subjects are randomly rematched into pairs. For each treatment we have data of
six independent observations collected with z-Tree (Fischbacher (2007)) at the University of
London. Altogether, 192 subjects participated in the experiment which lasted on average less
than an hour. Payoffs per period were exactly as shown in Figure 1 (with p = pence). Average
earnings were £11.28 (including a £5 show-up fee).
During the experiment, sellers’ track records were displayed in the left part of the screen
using a simple graphical representation. It showed four columns consisting of 30 hashes (#) -
each column representing one seller and each row representing one period. Initially, all
hashes were white. Then, after each period, hashes representing this period changed their
color. They turned black if the seller had not to make a decision because his buyer did not buy
from him. They turned green if the seller provided good quality of the good. They turned red
if the seller exploited the buyer’s trust and only provided bad quality. Additionally, in
2 This treatment reflects a network structure where buyers and sellers are linked in a way such that everybody
has full information.