Table 1: Overview of results
|
trust rate |
honor rate pe |
rformance rate | |
|
NO |
0.21 |
0.19 |
0.05 |
|
(0.16) |
(0.11) |
(0.05) | |
|
0.31 |
0.59 |
0.19 | |
|
DEGENERATE |
(0.14) |
(0.15) |
(0.12) |
|
0.44 |
0.48 |
0.23 | |
|
PARTIAL |
(0.17) |
(0.16) |
(0.14) |
|
FULL |
0.43 |
0.62 |
0.29 |
|
(0.15) |
(0.23) |
(0.18) | |
|
NO-DEGENERATE* |
p = 0.168 |
adjacent treatment effects, |
p = 0.005 |
|
DEGENERATE-PARTIAL* |
p = 0.064 |
p = 0.168 |
p = 0.344 |
|
PARTIAL-FULL* |
p = 0.468 |
p = 0.100 |
p = 0.235 |
|
NO-FULL* |
p = 0.027 |
further treatment effects p = 0.005 |
p = 0.008 |
|
NO-DEGENERATE- PARTIAL-FULL*__________ |
p = 0.007 |
p = 0.003 |
p = 0.001 |
Standard deviations are given in parentheses. Treatment effects are tested by one-tailed
Mann-Whitney U-tests (*) and Jonckheere-Terpstra tests (#) respectively.
Our design gradually increases the amount of feedback information from treatment to
treatment. As a consequence, several observed treatment differences of neighboring
treatments are sometimes insignificant (see Table 1). However, we observe a continuous
improvement in market performance due to higher network density. The Jonckheere-Terpstra
test shows that there is an ordering for all three key variables (trust rate, honor rate and
performance rate) of the treatments according to the network’s density.3
3 The Jonckheere-Terpstra test is a non-parametric test for ordered differences among classes. The alternative
hypothesis assumes a certain ordering of the medians of k statistically independent samples. All average values -
each of a statistically independent observation from a treatment with the same network density - are assigned to
one class.