Actions and Beliefs in a Trilateral Trust Game
Involving Germans, Israelis and Palestinians
Heike Hennig-Schmidt∗, Reinhard Selten †,
Gari Walkowitz‡, Eyal Winter§, & Ihab Dakkak¶
June 23, 2007
Abstract
We report results of an inter-cultural experiment between Germans, Israelis and
Palestinians based on a trust game. By eliciting players’ beliefs over the behavior
of their counterparts we are able to compare expectations with actual behavior.
We found levels of trust and reciprocity (beliefs) to differ significantly in the three
subject pools with high Palestinian and low Israeli amounts. The most startling
result of our experiment is the fact that players’ beliefs match well the actual behavior of their counterparts from their own country but they are completely wrong
in predicting the behavior of players in the other countries. Subjects attribute the
standard of behavior prevailing in their own environment also to subjects of the two
other countries. The most significant differences in interactions between individuals
of different cultures in our experiment are those between Israelis and Palestinians.
Israeli senders make low transfers (to all subjects) and Israeli responders’ beliefs
attribute low transfers to Palestinians. In contrast, Palestinian senders make high
transfers and Palestinian responders expect Israeli senders to make high transfers
as well. An interaction between a Palestinian sender and an Israeli responder is
likely to result in an outcome that will be perceived as positive discrimination while
an interaction between an Israeli sender and a Palestinian responder is expected to
create a false perception of negative discrimination.
JEL Classification: C7, C72, C81, C91, F00, O57
Keywords: trust, reciprocity, expectations, inter-cultural experiments, social
preferences
Acknowledgments: We thank Shani Kuna for valuable research assistance. We are grateful for
helpful comments and suggestions by participants of the following conferences: 3rd International Symposium
on Corporate Governance 2005, Nankai University, Tianjin, P.R. China; SERS 2006, Maria Laach, Germany;
Jahrestagung 2007, Sozialwissenschaftlicher Ausschuss, Verein für Socialpolitik, Magdeburg, Germany.
Financial support is gratefully acknowledged by Deutsche Forschungsgemeinschaft (SE 137/3-1).
∗
Bonn University
Bonn University
‡
Corresponding author: BonnE conLab, Laboratory for Experimental Economics, Bonn University, Adenauerallee 24-24, D-53113 Bonn, phone: +49-228-739196, fax: +49-228-719193, e-mail: [email protected].
§
Hebrew University of Jerusalem
¶
Bethlehem University
†
1
Introduction
The impact of social background and culture on individual behavior has been investigated in a
variety of studies in the recent experimental economics literature (e.g. Roth et al., 1991; Croson
et al., 1999; Anderson et al., 2000; Henrich, 2000; Henrich et al., 2001; Koford, 2001; Buchan et
al., 2004a,b; Cardenas and Carpenter, 2004; Lazzarini et al., 2004, Danileson and Holm, 2005;
Herrmann et al., 2007). Mainly, these papers explore similarities and differences among societies
by focusing on cross-cultural1 comparisons of behavior.
Only very few investigations consider interactive strategic decision situations in an intercultural context (Fershtman and Gneezy, 2001; Boarini et al., 2002; Willinger et al., 2003;
Bornhorst et al., 2004; Castro, 2007; Chuah et al., 2007; Goerg et al., 2007; Walkowitz et
al., 2007). The limited empirical evidence based on inter-cultural experiments is surprising,
in particular, because globalization leads to daily interactions between subjects from various
cultures holding specific social norms, values and behavioral attitudes. Culture-specific behavior
can generate mutual misunderstanding. Misinterpreting foreigners’ behavior in inter-cultural
exchange may lead to frictions and inefficiencies not primarily intended.
To investigate behavior in an inter-cultural multi-national environment we run an intercultural trust experiment (Berg et al., 1995) between Germans, Israelis and Palestinians2 , three
geographically distinct and culturally distant subject pools with different religious heritages
- Christianity for Germans, Judaism for Israelis and Islam for Palestinians. Moreover, the
historical and political background of the participating countries like Germany and Israel or
Israel and Palestine makes these subject pools a promising testbed for investigating culturally
induced behavioral differences.
By eliciting players’ beliefs over the behavior of their counterparts we are able to compare
expectations with actual behavior. We found levels of trust and reciprocity (beliefs) to differ
significantly in the three subject pools with high Palestinian and low Israeli amounts. The most
astonishing result of our experiment is the fact that players’ beliefs match well the actual behavior of their counterparts from their own country but they are completely wrong in predicting
the behavior of players in the other countries. Subjects attribute the standard of behavior prevailing in their own environment also to subjects of the two other countries thereby ignoring
cultural differences between countries (cf. Fukuyama 1995, 2001; Hofstede, 2001; Inglehart et
al., 2001). The most significant ones in our experiment are interactions between Israelis and
Palestinians. Israeli senders make low transfers (to all subjects) and Israeli responders’ beliefs
attribute low transfers to Palestinians. In contrast, Palestinian senders make high transfers and
Palestinian responders expect Israeli senders to make high transfers as well. An interaction
between a Palestinian sender and an Israeli responder is likely to result in an outcome that
1
We distinguish between intra-cultural, inter-cultural, and cross-cultural experiments. In intra-cultural experiments, participants from one country/culture interact with each other. Inter-cultural experiments investigate
behavior of subjects in different cultures interacting directly with each other. Cross-cultural experiments compare
behavior in two or more countries/cultures with subjects not interacting across cultures.
2
At the moment, a Palestinian state does not exist. Most of our Palestinian subjects are formally citizens of
Jordan and Israel. For the sake of simplifying the notation, we, nevertheless, refer to them as Palestinians.
1
will be perceived as positive discrimination while an interaction between an Israeli sender and a
Palestinian responder is expected to create a false perception of negative discrimination.
The remainder of this paper is organized as follows: In the next section, we describe the
design and methods of the trilateral trust game experiment. The procedure is explained in
detail in section 3. Section 4 presents our results on trust, reciprocity, beliefs and performance.
The final section 6 discusses our findings and concludes.
2
Experimental Design and Methods
The present trust experiment is a trilateral adaptation of the usual sequential protocol for
running investment game experiments (Berg et al. 1995). A sender and a responder first get
an endowment X=10 ECU (Experimental Currency Unit); the exchange rate was 1 ECU=0.50
US$ in each country3 . While the responder pockets his endowment the sender has to decide
on a transfer a ∈ [0, 1, ..., X − 1, X] to send to an anonymously matched responder of his own
or one of the two other countries. Each transfer a is tripled by the experimenter resulting in
the responder receiving 3a for each a. The responder then has to decide on a back-transfer
b ∈ [0, 1, ..., 3a − 1, 3a] to the sender. a can be interpreted as the sender’s trust towards the
responder. Likewise, b symbolizes the responders’ reciprocity towards the sender.
The game’s only sub-game perfect equilibrium is b∗ = a∗ = 0; the sender anticipating the
responder’s rational choice of b = 0 will transfer nothing (a = 0).
A trilateral multi-person multi-stage experiment like ours may encounter substantial organizational problems due to the separation of the three subject pools. The crucial constraint
is the interdependence of sequential choices at the three locations involved. The responder in
one country is not able to make her cross-border back-transfer without knowing the respective foreign-country sender’s transfer. We solve this problem by combining several experimental features known from the literature or developed by ourselves. We call our approach ELD
(E xperimenting over a Long Distance). ELD was first applied by Walkowitz (2004).
In the experiment reported here, ELD owns the following features.
1. Strategy method (Selten 1967). The strategy method allows to organizationally disconnect
the second stage of the game from the first stage. By having the responder state her
back-transfer for each possible transfer of the sender, the sequential two-person two-stage
game is converted into a two-person normal-form one-stage game for each player. These
correlated games can be played independently at different locations and different points
in time. For applications of the strategy method see also Brandts and Charness (2000),
Güth et al. (2003), and Bellemare et al. (2007).
2. Pen-and-paper. Using pen-and-paper makes the experimental design independent of equipment and software compatibility. Start-up costs are immensely reduced as the experiment
3
To avoid discrimination effects we did not adapt the exchange rate to local purchasing power. We paid a
locally adapted show up fee, however (5 US$ for G and I, and 3 US$ for P).
2
can be run in a non-laboratory environment like a classroom. This is particularly important
in our trilateral context where a common hardware and software standard was not available. Moreover, pen-and-paper creates transparency as to the experimental procedures
enhancing their credibility for participants not used to experimental methods.
3. Remote-control organization. We designed a sophisticated experimental procedure to ensure equivalent experimental conditions in all three subject pools. A central unit, the
Chief Experimenter (CE) was responsible for overall planning and controlling the experiment. CE was located in Bonn. Local experimenters (LEs) were in charge of organizing
and running the sessions in Jerusalem and Bethlehem4 . CE instructed all LEs in advance by distributing a detailed procedural script and an extensive instruction manual in
English also checking LEs’ understanding of the rules of the game. CE prepared identical experimental instructions and decision sheets. The translation from German into
the local languages was done in Bonn by native speakers applying the back-translation
method (Brislin 1970, Eco 2000). CE ex-ante coded and prepared the decision sheets and
randomly matched players across subject pools. All material was shipped to LEs well in
advance of running the experiment. Participants at each location randomly draw a personal identification code constituting a predefined order of matching across subject pools,
not noticeable for participants. The code also ensured full anonymity by a double-blind
procedure. Subjects then made their choices on decision sheets marked with their code
number and displaying their counterpart’s pool affiliation. All sessions in all locations
having been finished, CE collected the data, computed the payoffs and transferred this
information to all LEs. Finally, subjects were paid out by LEs two to three days after the
last session in any of the subject pools has been finished.
3
Experimental Procedure
The trilateral experiment was run simultaneously at BonnE conLab (Laboratory for Experimental Economics,University of Bonn, Germany), at RatioLab (Laboratory for the Study of Interactive Decision Making,Hebrew University of Jerusalem, Israel), and at Bethlehem University,
Palestine. 90 students participated mayoring in different disciplines at the above universities.
Subjects were recruited by campus advertisements5 promising a monetary reward for participation in a decision-making task. We strictly controlled for participants’ nationalities in order to
exclude foreign players from national subject pools6 . In each of the three countries, we ran a
sender session and a responder session with 15 participants each7
4
In inter-cultural research, it is advisable to work with local experimenters in each national subject pool
order to avoid self-presentation and face-saving effects (Bond and Hwang, 1986) as well as possible effects
foreign experimenters.
5
In Germany, subjects were recruited by the online recruiting system ORSEE (Greiner 2004).
6
In Jerusalem, Arabic Jews did not participate in the experiment.
7
We had to restrict the number of independent observations to 15 due to limited availability of subjects
Bethlehem University. Moreover, we confined our analysis to one university per country as we were not able
3
in
of
at
to
Any German, Israeli or Palestinian participant played the trust game with one member of
his/her own country as well as with one member of the other two countries. All matchings were
played one-shot. Thus, each player interacted with one subject of any of the three national
subject pools once, making choices in three subsequent decision rounds. In each round, any
player was endowed with X=10 ECU with each sender making a transfer a and each responder
deciding on a back-transfer b. Participants kept their player type throughout the experiment. A
player’s total payoff was the sum of the payoffs over all three decision rounds plus the show-up
fee.
When introducing the experiment, we informed subjects about the countries involved in
the study by providing them with the names of the participating universities and the countries
they are located in. We chose this procedure to guarantee a complete comprehension of the
experimental setup. Due to religious, organizational and technical constraints, the experiment
could not be run at exactly the same time at all locations. We, therefore, conducted sessions on
three consecutive days. At each location, LEs ran the sender and responder sessions within one
day.
The instructions of the trilateral trust game were read aloud and explained in detail to the
experimental subjects.8 Questions were asked and answered in private. We took great care
to ensure participants’ understanding the game and the underlying incentive structure. After
having read the instruction, for which they had plenty of time, subjects had to answer four
control questions testing their understanding of the decision situation. We did not proceed in
the experimental protocol until all subjects had answered all questions correctly. Thus, we can
safely assume that people fully understood the game, its features and incentives.
After having answered the control questions, senders had to make their transfer choice a and
responders had to write down their back-transfer decision b for all potential values of a. For
each decision round, subjects were provided with a separate decision sheet to inhibit changing
prior decisions or recouping information from previous choices. Having made all three decisions
responders were asked to guess senders’ transfers a, and senders were asked to guess responders’ back-transfers b for every potential value a. When making their transfer or back-transfer
decisions, subjects did not know about the subsequent estimation task.9
Participants played the game against a counterpart from each subject pool without being
informed about their income before the very end of the entire experiment. Subjects took their
decisions in complete anonymity. All participants were fully informed on all features of the
experimental design and the procedures. Sessions lasted for about 90 minutes including reading
the instructions. On average, subjects earned 23.9 US$ (Germans 23.2 US$, Israelis 24.6 US$,
and Palestinians 24.0 US$).
find a second Palestinian university to be included into our study.
8
See Appendix A for the English translation of the instruction. Instructions in German, Hebrew and Arabic
are available from the authors upon request.
9
We decided on this timing of events because we wanted to avoid subjects’ decisions being influenced by the
subsequent belief elicitation.
4
4
Results
We first analyze senders’ transfers and responders’ beliefs on these transfers in subsection 4.1.
Subsequently, we focus on responders’ back-transfers and senders’ beliefs on these back-transfers
(subsection 4.2). We also contrast actions and beliefs of participants from different countries.
Finally we study payoffs and performance (subsection 4.3.)
We use abbreviations G, I, and P for identifying Germans, Israelis, and Palestinians. In
general, we analyse the data according to the alphabetical order of the country names. We
denote senders’ transfers also as trust and responders’ back-transfers as reciprocity. As we have
no specific behavioral hypotheses, all statistical tests are two-sided.
4.1
Senders’ transfers a and responders beliefs ã
We start our analysis by investigating trust behavior (senders’ transfers a) and responders’
beliefs ã on senders’ trust. Table 1 shows average a and ã from and towards each of the three
countries separately (columns II to V and lines 1 to 6). Column VI and lines 7 and 8 display
aggregated transfers and beliefs from and towards all countries. Numbers in parantheses are
standard deviations.
Population
Measure
To G
To I
To P
To all countries
(I)
(II)
(III)
(IV)
(V)
(VI)
(1)
(2)
(3)
(4)
(5)
(6)
From G
a
ã
a
ã
a
ã
4.4
4.8
3.9
4.8
6.9
6.5
(7)
(8)
From all countries
a
ã
5.1
5.4
From I
From P
(3.3)
(3.1)
(3.1)
(3.4)
(1.6)
(2.5)
(3.0)
(3.1)
5.2
4.5
3.9
4.2
6.2
6.9
5.1
5.2
(3.5)
(3.5)
(3.4)
(3.3)
(2.5)
(2.7)
(3.3)
(3.4)
5.6
4.7
2.9
3.7
6.7
7.7
5.1
5.4
(3.3)
(3.3)
(3.4)
(3.6)
(2.3)
(2.7)
(3.4)
(3.6)
5.1
4.7
3.6
4.2
6.6
7.0
5.1
5.3
(3.3)
(3.2)
(3.3)
(3.4)
(2.1)
(2.6)
(3.2)
(3.3)
Table 1: Senders’ average transfers a and responders’ average beliefs ã on transfers a across countries
4.1.1
Transfers a
According to predictions of standard game theory, we should observe a∗ = 0. In line with the
literature (e.g. Berg et al., 1995), we find subjects in all subject pools to invest considerable
amounts, however. Aggregated over all countries, 5.1 ECU are transferred to responders, with
median = 5, mode = 5 and sd = 3.2. In G, 86.7% of the senders transfer a positive amount, in
I the percentage is 66.7%, and in P it is even 100%; full trust is displayed most frequently in G
(17.8%), followed by P (13.3%) and I (6.7%) (see figure 1a).
Thus, the standard game theoretic prediction is to be rejected for all countries and for all
matchings. It is remarkable that subjects show substantial trusting behavior in the inter-cultural
5
context, in particular given the historical and political background of the participating countries
like G and I or I and P.
We next analyse trust in the participating subject pools. To this end, we make a crosscultural comparison of aggregated trust behavior by summing up all transfers from a specific
subject pool to the responders of all three countries (lines 1, 3, 5 and column VI in table 1).
b) Responders’ beliefs on a
30
25
20
15
10
0
5
Frequency (percentage)
35
40
a) Senders’ transfers a
0
1
2
3
4
5
6
7
8
9
10
0
1
2
3
4
5
6
7
8
9
10
Transfer amount a
Germans
Israelis
Palestinians
Graphs by transfers_and_beliefs
Figure 1a, b: a) Senders’ transfers a b) Responders’ beliefs ã per country
Result T110 : Aggregated trust is different in all three countries.
SUPPORT: Aggregated transfers are significantly different across countries (p=0.000, KruskalWallis test). Highest transfers are made by P (6.6 ECU), followed by G (5.1 ECU), with lowest
trust being shown by I (3.6 ECU) yielding the relation aP > aG > aI . A pairwise crosscultural analysis of individual transfers shows highly significant differences as well, with aG > aI
(p=0.040), aP > aG (p=0.022) and aP > aI (p=0.000) (Mann-Whitney-U test).
Our previous analysis showed different levels of trust to exist in the participating subject
pools. The aggregated data do not allow to observe potential inter-cultural dicrimination,
though. We therefore take a closer look at transfers to specific subject pools (lines 1, 3, 5
and columns III, IV, V in table 1).
Result T2: I- and P-senders do no distinguish between subject pools. G-senders transfer less
to fellow-country responders than to P-responders.
SUPPORT: G on average transfer most to P (5.6 ECU), second most to I (5.2 ECU), and least
to G (4.4 ECU) thereby discriminating compatriots against P (p=0.040, Wilcoxon-signed-ranks
test). I, on average, show equal trust to G and I (3.9 ECU). They transfer only 2.9 ECU to P,
however, the smallest average transfer in our study (see table 1). P, on the other hand, show
10
T denotes T rust.
6
the highest trusting behavior of all three countries transferring on average 6.9 ECU to G, 6.7
ECU to P, and 6.2 ECU to I. A Friedman test does not indicate any discrimination of I-senders
against or in favor of one country (p=0.308); the same holds for P (p=0.582).
The results so far are different levels of trust between subject pools and practically no
discrimination. The interplay of these two factors might result in dramatic effects in direct
interaction, however. We, therefore, now analyse bilateral interactions within the trilateral
context. We compare direct transfers between countries, for instance G-transfers to I (line 1,
column IV in table 1) and I-transfers to G (line 3, column III in table 1).
Result T3: I and P face a significant discrepancy of mutual trust with low I- and high Ptransfers.
SUPPORT: On average, G transfer to I (P) 5.2 ECU (5.6) whereas I (P) transfer to G 3.9
(6.9). The differences between G and I(P) turn out not to be significant (p=0.292, and p=0.193,
respectively, Mann-Whitney-U test). I and P, however, face a huge discrepancy of mutual trust
with I on average transferring 2.9 ECU11 to P and P transferring 6.2 ECU to I. This difference
is highly significant (p=0.008, Mann-Whitney-U test).
The dramatic difference in bilateral trust, in particular between I and P, might create a
false perception of the intentions underlying the transfer decisions if subjects are not aware of
the different trust levels across the societies. We, therefore, next investigate beliefs ã. It is
especially interesting whether beliefs on trust in foreign subject pools match beliefs on trust
of fellow-participants. Remember that it is the responders who make the belief statements on
senders’ trust.
4.1.2
Beliefs ã
In line with our results on transfers, we find responders to expect substantial trust behavior.
Aggregated over all countries responders believe that on average 5.3 ECU will be transferred,
with median = 5, mode = 10 and sd = 3.3. In G and I, 86.7% of the responders believe in
a positive transfer, in P the percentage is 100%; full trust is most frequently expected by P
(28.9%), followed by G and I (each 13.0%), (see figure 1b).
Thus, responders in all countries and matchings do not expect the standard game theoretic
prediction of zero-transfers. Note responders’ expectations of substantial high trusting behavior
also in the inter-cultural context despite the historical and political background of the participating countries.
We next analyse beliefs on trust in the participating subject pools. To this end, we compare
aggregated beliefs ã across cultures by summing up all beliefs on transfers from a specific subject
pool to the responders of all three countries (lines 2, 4, 6 and column VI in table 1).
Result T4 : Aggregated beliefs on senders’ trust are different across countries.
11
This is the lowest country-specific transfer in our experiment.
7
SUPPORT: Aggregated beliefs ã are significantly different across countries (p=0.000, KruskalWallis test). Highest beliefs are stated by P (7.0 ECU), second highest by G (4.7 ECU),
and lowest by I (4.2 ECU). Thus, the same relation as for transfers also holds for beliefs,
namely ãP > ãG > ãI . A pairwise cross-cultural analysis of individual beliefs shows significant
differences, with ãP > ãG (p=0.000) and ãP > ãI (p=0.000). G- and I-beliefs do not differ
significantly (p=0.600) (all Mann-Whitney-U test).
Our previous analysis showed different levels of expectations on trust to exist in the participating subject pools. Yet, the aggregated data do not allow to observe potential inter-cultural
differences. We therefore continue with investigating the beliefs responders hold on senders’
transfers towards specific subject pools (lines 2, 4, 6 and columns III, IV, V in table 1).
Result T5: Responders in G and P do not differentiate in beliefs. I-responders expect lower
transfers from P- than from G-senders.
SUPPORT: G(P)-responders’ expect on average 4.5, 4.7, 4.8 (6.5, 6.9, 7.7) ECU (see table 1)
from their counterparts. I-responders on average expect lower transfers from P-senders (3.7
ECU) than from G- and I-senders (4.8 ECU and 4.2 ECU, respectively). A Friedman test
shows that G- and P-responders do not differentiate between subject pools (G: p=0.733; P:
0.665) whereas I-responders do (p=0.080) expecting higher transfers from G-senders than from
P-senders (p=0.031, Wilcoxon-signed-ranks test).
As with trust, we observe different levels of beliefs on trust and basically no differentiation
in the three subject pools. Next, we analyse mutual beliefs by comparing bilateral expectations
on senders’ transfers.
Result T6 : I and P face a significant discrepancy of mutual beliefs on transfers, with low Iand high P- expectations.
SUPPORT: As for transfers, a bilateral comparison of beliefs about transfers shows broad discrepancies between I and P. While beliefs of G and I as well as of G and P do not significantly differ (p=0.795 and p=0.117, respectively, Mann-Whitney-U test), beliefs of I and P do
(p=0.019, Mann-Whitney-U test). I-responders expect a transfer of 3.7 ECU from P-senders
being the lowest overall belief on transfers in the experiment. P-responders, on the other hand,
expect I-senders to back-transfer 6.9 ECU.
4.1.3
Comparison of transfers a and beliefs ã
The most important results so far are practically no discrimination in trust between countries
but huge differences in the levels of transfers in all three subject pools with a dramatic discrepancy between I and P. The same applies for beliefs. A false perception of negative (positive)
discrimination might, however, arise if subjects are ignorant about the divergence in trust between countries. It has been shown in the literature that a mismatch of choices and expectations
can lead to aggressive behavior e.g. in sequential interactions (Bosman and van Winden, 2002;
Bosman et al., 2006). We are able to check for subjects’ awareness of the differences in trust by
contrasting senders’ actual transfers a and responders’ beliefs ã.
8
In a first step, we compare transfers and beliefs within a country (diagonal of table 2). This
comparison gives important insights as a correspondence of transfers and beliefs would point to
the existence of trust norms within this country (see e.g. Parsons, 1952), and a difference between
countries would indicate the existence of different country-specific trust norms. Remember that
senders’ transfers and responders’ beliefs on these transfers are independently elicited from
different experimental subjects being members of different experimental sessions.
(1)
(2)
(3)
Population
G ã
I ã
P ã
(I)
(II)
(III)
(IV)
Ga
Ia
Pa
4.4
3.9
6.9
(3.3)/4.8 (3.1)
(3.1)/4.5 (3.5)
(1.6)/4.7 (3.3)
5.2
3.9
6.2
(3.5)/4.8 (3.4)
(3.4)/4.2 (3.3)
(2.5)/3.7 (3.6)
5.6
2.9
6.7
(3.3)/6.5 (2.5)
(3.4)/6.9 (2.7)
(2.3)/7.7 (2.7)
Table 2: Average country-specific transfers a and beliefs ã
Result T7: Responders’ beliefs match well senders’ actual behavior from their own country.
SUPPORT: Comparing senders’ transfers and responders beliefs country-wise, we find no statistical difference (G: p=0.812, I: p=0.910, P: p=0.162, Mann-Whitney-U test). There is no
evidence in the data that responders expect different amounts than senders from their own country actually transfer. This is a striking result given that transfers and beliefs are derived from
individuals deciding independently of each other.
A further important question is how well responders predict foreign senders’ behavior. If
they, for instance, attribute the standards of their own country also to foreign subject pools,
direct interactions might results in substantial surprises because of the cross-cultural differences
in trust reported above. To answer this question, we compare senders’ actual within-country
transfers with the beliefs fellow-country respondors hold towards foreign-country senders. To
give an example for G-participants, we compare intra-cultural G-transfers of 4.4 [row (1), column
(I) in table 2] with G-beliefs on I-transfers [4.5, row (2), column (I)], and P-transfers [4.7, row
(3), column (I) in table 2].
Result T8: Responders attribute the transfer standards of their own population also to foreign
subject pools.
SUPPORT: G-senders on average transfer 4.4 ECU to G-responders while G-responders expect
4.5 ECU from I-senders and 4.7 ECU from P-senders. I-senders on average send 3.9 ECU to
I-responders; I-responders believe G-senders to send 4.8 ECU and P-senders to send 3.7 ECU.
P-senders on average transfer 6.7 ECU to P-responders while P-responders believe G-senders
to transfer 6.5 ECU and I-senders to transfer 6.9 ECU. We find no statistical difference between
transfers and beliefs in the different matchings (G vs. I: p=0.977; G vs. P: p=0.812; I vs. G:
p=0.486, I vs. P: p=0.892; P vs. G: p=0.926, P vs. I: p=0.961, Mann-Whitney-U test).
We finally analyze how well responders predict senders’ behavior in the other countries. To
this end, we bilaterally compare country-specific transfers a to a specific population with the
expectations ã this population holds (see table 2).
9
Result T9 : Responders are remarkably wrong in predicting senders’ behavior in the other
countries, in particular when P is involved.
SUPPORT: A bilateral comparison of transfers and beliefs in the I/P-matching shows a dramatic
discrepancy. For one thing, I-senders on average transfer 2.9 ECU to P-responders while Presponders expect 6.9 ECU. This difference is highly significant (p=0.002, Mann-Whitney-U
test). On the other hand, P-senders transfer 6.2 ECU to I-responders who expect a significantly
lower amount of only 3.7 ECU (p=0.037, Mann-Whitney-U test). Furthermore, P slightly
outperfom G-expectations (4.7 ECU) with an average transfer of 6.9 to G (p=0.055, MannWhitney-U test). There appears to be no difference in the other interactions involving G,
though. On the one hand, G-transfers do not mismatch foreign expectations. I on average
expect 4.8 ECU from G and receive 5.2 ECU (p=0.815, Mann-Whitney-U test). P expect an
average G-transfer of 6.5 ECU and receive 5.6 ECU (p=0.490, Mann-Whitney-U test). On the
other hand, I meet G-expectations. G, on average expect a transfer of 4.5 ECU and actually
receive 3.9 ECU from I (p=0.672, Mann-Whitney-U test).
4.2
Responders’ back-transfers b and senders’ beliefs b̃
In this subsection, we investigate reciprocity behavior (responders’ back-transfers b) and senders’
beliefs b̃ on responders’ reciprocity. We base our analysis on the full response vectors gained by
the strategy method. But instead of reporting the absolute values, in the following we state b
(b̃) as a percentage of 3a, (3ã) for a(ã) > 0. Table 3 shows average percentage b and b̃ from and
towards each of the three countries separately (columns II to V and lines 1 to 6). Column VI
and lines 7 and 8 display aggregated back-transfers and beliefs from and towards all countries.
Numbers in parantheses are standard deviations.
Population
Measure
to G
(I)
(II)
(1)
(2)
(3)
(4)
(5)
(6)
from G
b
b̃
b
b̃
b
b̃
40.3
36.1
37.0
41.0
56.9
75.6
(22.3)
(7)
(8)
from all countries
b
b̃
44.8
50.9
(29.3)
from I
from P
(III)
(23.0)
(30.4)
(23.2)
(30.6)
(19.2)
(28.0)
to I
(IV)
38.6
36.2
40.1
40.3
51.1
66.2
(21.8)
43.3
47.6
(28.9)
(23.8)
(30.7)
(25.3)
(31.6)
(25.3)
(28.1)
to P
to all countries
(V)
41.1
39.9
35.6
28.0
49.1
74.2
(20.4)
41.9
47.4
(28.8)
(21.3)
(32.5)
(26.2)
(30.4)
(24.4)
(31.0)
(VI)
40.0
37.4
37.6
36.4
52.4
72.0
(21.5)
43.3
48.6
(29.0)
(22.7)
(31.2)
(25.6)
(31.0)
(23.4)
(29.1)
Table 3: Responders’ average percentage back-transfers b and
senders’ average percentage beliefs b̃ on back-transfers b across countries
4.2.1
Back-transfers b
According to predictions of standard game theory, we should observe b∗ = 0. In line with
the literature (e.g. Berg et al., 1995), we find subjects in all subject pools to back-transfer
considerable amounts, however. In 83% of all cases where b > 0, responders made a positive
10
back-transfer. Aggregated over all countries, a fraction of 43.3% is back-transferred to senders,
with median=44.4%, mode=0 and sd=29.0.
In G and I, responders backtransfer 37.0%(30.4%) to 41.3%(41.1%) with reciprocity growing
when transfers a increase. P, however, backtransfer 56.3% to 49.6% with reciprocity declining
when transfers a increase (see figure 2a).
b) Senders’ beliefs on b
70
60
50
40
30
20
Back−transfer b (percentage)
80
a) Responders’ back−transfers b
0
3
6
9
12 15 18 21 24 27 30
0
3
6
9
12 15 18 21 24 27 30
Received amount 3a
Germans
Israeli
Palestinians
Graphs by transfers_and_beliefs
Figure 2a, b: a) Responders’ average percentage back-transfers b and
b) senders’ average percentage beliefs b̃ on back-transfers b across countries
We next analyse reciprocity in the participating subject pools. To this end, we make a crosscultural comparison of aggregated reciprocity behavior by looking at all back-transfers from a
specific subject pool to the senders of all three countries [lines 1, 3, 5 and column (VI) in table
3].
Result R112 : Aggregated reciprocity is different in all three countries.
SUPPORT: Aggregated back-transfers are significantly different across countries (p=0.000,
Kruskal-Wallis test). Highest back-transfers are made by P (52.4%), followed by G (40.0%)
with lowest reciprocity being shown by I (37.6%). This yields the relation bP > bG > bI in line
with our findings on senders’ trust behavior. A pairwise cross-cultural analysis of individual
back-transfers shows highly significant differences as well, with bG > bI (p=0.002), bP > bG
(p=0.000), and bP > bI (p=0.000)(all Mann-Whitney-U test).
Our previous analysis showed different levels of reciprocity to exist in the participating
subject pools. The aggregated data do not allow to observe potential inter-cultural dicrimination,
though. We therefore take a closer look at back-transfers to specific subject pools (lines 1, 3, 5
and columns III, IV, V in table 3).
Result R2: Responders show different reciprocity behavior towards specific subject pools: G
do no distinguish between countries. I back-transfer less to G- and P- than to fellow-country
senders. P favor G- against I- and P-senders.
12
R denotes Reciprocity.
11
SUPPORT: G-responders do not differentiate across countries, back-transferring 41.1% to P,
40.3% to G and 38.6% to I (p=0.647, Friedman-test). A pairwise comparison confirms that Gresponders do not act in favor of or against any national group (p=0.173 for G and I, p=0.911
for G and P, and p=0.745 for I and P, all Wilcoxon-signed-ranks test). I-responders, however,
do distinguish according to senders’ origin (p=0.003, Friedman-test). They back-transfer most
to their compatriots (40.1%), 37.0% to G and least to P (35.6%) thereby discriminating G and
P against I (p=0.002, p=0.000, respectively, Wilcoxon-signed-ranks test). I do not substantially differentiate between G and P (p=0.172, Wilcoxon-signed-ranks test). P-responders also
discriminate (p=0.000, Friedman-test). While they favor G (56.9%) compared to I (51.1%) and
their compatriots (49.1%) (p=0.007, p=0.000, respectively, Wilcoxon-signed-ranks test) I- and
P-senders are not treated differently.
The results so far are different levels of reciprocity between subject pools and some tendencies
for discrimination in I and P. The interplay of these two factors might result in dramatic effects
in direct interaction. We, therefore, now analyse bilateral exchanges by comparing direct backtransfers between countries, for instance, G-back-transfers to I (line 1, column IV in table 1)
and I-back-transfers to G (line 3, column III in table 3).
Result R3: As to mutual reciprocity, G- and I-responders make substantially lower backtransfers than P-responders.
SUPPORT: On average, G back-transfer nearly as much to I (38.6%) as I send back to G
(37.0%) (p=0.133, Mann-Whitney-U test). Yet, G and P as well as I and P face a significant
discrepancy with P back-transferring 56.9% (51.1%) to G (I) and G (I) returning only 41.1%
(35.6%) (both p=0.000, Mann-Whitney-U test).
The huge differences in bilateral reciprocity when interacting with P might create a false
perception of the intentions for back-transfer decisions, in particular when subjects are not
aware of the different reciprocity levels in all three societies. We, therefore, next investigate
beliefs b̃. Do beliefs on reciprocity of fellow-participants match beliefs on reciprocity in foreign
subject pools? Remember that senders make the belief statements on responders’ reciprocity.
4.2.2
Beliefs b̃
In this subsection, we analyse sender’s beliefs b̃ on responders’ back-transfers b . We again use
the complete dataset on response vectors gained by the strategy method. Beliefs are calculated
like back-transfers in subsection 4.2.1.
In line with our results on back-transfers, we find senders to expect substantial positive
reciprocity behavior. In 88.7% of all cases where a > 0, senders anticipate a positive backtransfer believing that on average a fraction of 48.6% is back-transferred, with median = 45.8%,
mode = 33.3% and sd = 29.1%. In G (I), senders expect an average back-transfer of 33.3%
(37.0%) to 39.7% (32.5%) with reciprocity growing (decreasing) with increasing a. P, however,
expect an average backtransfer of 63.0% to 76.8% with reciprocity rising in transfers a (see figure
2b).
12
Thus, senders in all countries and matchings do not expect the standard game theoretic
prediction of zero back-transfers. Note senders’ expectations of substantial high reciprocity
behavior also in the inter-cultural context.
We next analyse beliefs on reciprocity in the participating subject pools. To this end, we
make a cross-cultural comparison of aggregated beliefs b̃ (lines 2, 4, 6 and column VI in table 3)
as we did with back-transfers b in section 4.2.1.
Result R4 : Aggregated beliefs on responders’ reciprocity are different in all three countries.
SUPPORT: Aggregated beliefs b̃ are significantly different across countries (p=0.000, KruskalWallis test). Highest beliefs on back-transfers are stated by P (72.0%). G and I expect a
back-transfer of 37.4% and 36.4%, respectively. Thus the same relation b̃P > b̃G > b̃I holds for
beliefs b̃ as for back-transfers b. A pairwise cross-cultural analysis of individual beliefs shows
significant differences, with b̃G > b̃I , b̃P > b̃G , and b̃P > b̃I (p=0.004, p=0.000, p=0.000,
respectively, all Mann-Whitney-U test).
Our previous analysis showed different levels of expectations on reciprocity to exist in the
participating subject pools. Yet, the aggregated data do not allow to observe potential intercultural differences. We, therefore, now analyse the beliefs senders hold on responders’ backtransfers towards specific subject pools (lines 2, 4, 6 and columns III, IV, V in table 3).
Result R5: Senders hold different beliefs on specific subject pools’ reciprocity behavior: Gsenders expect higher back-transfers from P- than from G- and I-responders. I-senders anticipate
higher reciprocity from G- and I- than from P-responders. P-senders believe in higher backtransfers from G- and P- than from I-responders.
SUPPORT: G-senders discriminate in their beliefs expecting higher back-transfers from P(39.9%) than from G- and I-responders (36.1% and 36.2%, respectively) (p=0.000, Friedmantest). I-senders believe to get back significantly lower amounts from P-responders, however,
(28.0%) than from G- (41.0%) and I-responders (40.3%) (p=0.000, Friedman-test). Likewise,
P-senders expect significantly lower back-transfers from I-responders (66.2%) than from G(75.6%) and P-responders (74.2%) (p=0.000, Friedman-test).
As with reciprocity, we observe different levels of beliefs on reciprocity and a tendency to
discriminate between subject pools. Next, we analyse mutual beliefs by comparing bilateral
expectations on responders’ back-transfers.
Result R6: G- and P- as well as I- and P-senders face a significant discrepancy of mutual
beliefs on reciprocity, with low G- and I- and high P-expectations.
SUPPORT: As for back-transfers b, a bilateral comparison of b̃ shows no significant difference
between G and I (p=0.940, Mann-Whitney-U test). Yet, there are broad discrepancies when P
is involved. While P expect an average back-transfer of 75.6% from G, G expect to get back
only 39.9% from P (p=0.063, Mann-Whitney-U test). I expecting to receive 28.0% back from
P are believed to back-transfer 66.2% to P (p=0.034, Mann-Whitney-U test).
13
4.2.3
Comparison of back-transfers b and beliefs b̃
The most important results so far are some tendencies of discrimination in reciprocity between
countries and substantial differences in the levels of back-transfers in all three subject pools. For
beliefs, more or less the same applies. A false perception of negative (positive) discrimination
might, however, arise if subjects are ignorant about the divergence in reciprocity between countries. We are able to check for subjects’ awareness of this difference by contrasting responders’
actual back-transfers b and senders’ beliefs b̃.
In a first step, we compare back-transfers and beliefs within a country (diagonal of table 4).
As with transfers, a correspondence of back-transfers and beliefs would point to the existence
of reciprocity norms within this country and a difference between countries would indicate the
existence of different country-specific reciprocity norms. Remember that back-transfers b and
beliefs b̃ are independently elicited from responders and senders.
Result R7: In G and I, senders’ beliefs match well responders’ actual behavior from their own
country. P-senders, however, overestimate reciprocity behavior of fellow-country responders.
SUPPORT: Comparing G-responders’ back-transfers (40.3%) and G-senders’ beliefs (36.1%),
there is no statistical difference (p=0.123, Mann-Whitney-U test). The same holds for I-senders (
40.1%) and I-responders (40.3%) (p=0.313, Mann-Whitney-U test). P-senders, however, expect
significantly higher back-transfers (74.2%) from fellow-country responders than these actually
return (49.1%) (p=0.000, Mann-Whitney-U test).
(1)
(2)
(3)
Population
Gb̃
Ib̃
Pb̃
(I)
(II)
(III)
(IV)
Gb
Ib
Pb
40.3
37.0
56.9
(22.3)/36.1 (23.0)
(30.4)/36.2 (23.8)
(30.6)/39.9 (21.3)
38.6
40.1
51.1
(21.8)/41.0 (23.2)
(30.7)/40.3 (25.3)
(31.6)/28.0 (26.2)
41.1
35.6
49.1
(20.4)/75.6 (19.2)
(31.2)/66.2 (25.3)
(30.4)/74.2 (24.4)
Table 4: Average country-specific back-transfers b and beliefs b̃
A further important question is how well senders predict foreign responders’ behavior. If
they, for instance, attribute the standards of their own country also to foreign subject pools, direct interactions might results in substantial surprises because of the cross-cultural differences in
reciprocity reported above. To answer this question, we first compare responders’ actual withincountry back-transfers with the beliefs fellow-country senders hold towards foreign-country responders. To give an example for P-participants, we compare intra-cultural P-back-transfers
of 49.1% [row (3), column (IV) in table 4] with P-beliefs on G-back-transfers [75.6%, row (1),
column (IV)], and I-back-transfers [66.2%, row (2), column (IV)].
Result R8: No significant differences can be detected for G. I back-transfer relatively higher
amounts to I than I-senders expect to receive from P. Moreover, P state significantly higher
beliefs towards G- and I-responders than own-country responders actually back-transfer.
SUPPORT: We find substantial differences between transfers and beliefs in the different matchings. G-responders on average back-transfer 40.3% to G-senders while G-senders expect 36.2%
14
from I-responders and 39.9% from P-responders (p=0.553 and p=0.000, respectively, MannWhitney-U test). I-responders on average back-transfer 40.1% to I-senders while I-senders
believe G-responders to back-transfer 41.0% and P-responders to back-transfer 28.0% (p=0.056
and p=0.000, respectively, Mann-Whitney-U test). P-responders on average back-transfer 49.1%
to P-senders while P-senders believe G-responders to back-transfer 75.6% and I-responders to
back-transfer 66.2% (p=0.000 and p=0.000, respectively, Mann-Whitney-U test).
We finally analyze how well senders predict responders’ behavior in the other countries. To
this end, we bilaterally compare country-specific back-transfers b to a specific population with
the expectations b̃ this population holds (table 4).
Result R9: Senders are remarkably wrong in predicting responders’ behavior in the other countries, in particular when P is involved.
SUPPORT: A bilateral comparison of back-transfers and beliefs in the G(I)/P-matching shows
a dramatic discrepancy. For one thing, G(I)-responders on average back-transfer 41.1%(35.6%)
to P-senders while P-senders expect 75.6%(66.2%). These differences are highly significant
(both p=0.000, Mann-Whitney-U test). Moreover, P-responders back-transfer 56.9%(51.1%)
to G(I)-senders who expect a significantly lower back-transfers of only 39.9%(28.0%) (both
p=0.000, Mann-Whitney-U test). There appear to be only slight differences when G and I
are involved. On the one hand, G-back-transfers (38.6%) only weakly mismatch I expectations
(41.0%) (p=0.056, Mann-Whitney-U test). On the other hand, I (37.0%) meet G-expectations
(36.2%) (p=0.553, Mann-Whitney-U test) (see table 7).
4.3
Performance
In this section, we investigate participants’ actual performance. To this end, we study senders’
and responders’ payoffs (excluding the show-up fee). We base our analysis of responders’ performance on their actual back-transfers b̄ as a percentage of 3a. This is different from what we
did in section 4.2 where we based our analysis on responders’ full response vectors.13
On average, responders make significantly higher payoffs (30.4 US$) than senders (17.4 US$)
(p=0.000, Mann-Whitney-U test) (table 5). This is due to the fact that on average responders
do not share the total pie (2X + 2a) equally and back-transfer b̄ < 2a, i.e. less than two thirds
of 3a.
When aggregating senders’ and responders’ payoffs, I achieve highest amounts (24.6 US$),
followed by P (24.0 US$) and G (23.2 US$). Payoff levels do not differ across countries (p=0.900,
Kruskal-Wallis test).
Players
G
I
P
all
(I)
(II)
(III)
(IV)
(V)
(1)
(2)
Senders
Responders
17.6
28.7
17.4
31.7
17.2
30.8
17.4
30.4
(3)
all
23.2
24.6
24.0
23.9
13
Highest actual back-transfers b̄ are made by P (53.1%) followed by G (41.0%) and I (39.0%). Like with b we
have the relation b̄P > b̄G > b̄I .
15
Table 5: Senders’ and responders’ average payoffs ($)
Did trust pay in the trilateral trust game? On average, absolute amounts of back-transfers
increase in a, implying that more trust is rewarded more in absolute terms in all matchings. For
relative rewarding we observe a different picture. While G and I reward higher transfers a with
higher back-transfers b̄, P do the opposite. The larger the amount 3a P-responders receive, the
smaller is the back-transfer b̄. See also figure 2a for back-transfers b based on strategy vectors.
Moreover, even though in G and I b̄ increases in a reciprocity levels are much lower than that
of P. Discriminating rewarding patterns in favor or against a specific country can not be found.
Finally, we calculate the rate of return R, i.e. the percentage a sender receives back in
addition to his investment, a measure indicating how well trust pays off14 . Over all matchings,
trusting P-responders yields the highest return R for all senders. This holds in particular for Gand I-senders getting 73.6% (62.3%) back in addition to their transfers to P (table 6). Trusting
I, however, is a bad deal for P- and G-senders as they realize the lowest (second lowest) return
of 12.8% (16.9%). For I-senders, it turns out to be a bad bargain to trust G because of getting
a return of only 17.0%.
Population
to G
to I
to P
to all countries
(I)
(II)
(III)
(IV)
(V)
(1)
(2)
(3)
from G
from I
from P
24.1
17.0
23.8
16.9
30.0
12.8
73.6
62.3
44.5
40.0
35.0
27.0
(4)
from all countries
22.0
18.8
59.4
33.4
Table 6: Average rates of return in %.
5
Discussion and Conclusion
We have reported experimental results of an inter-cultural trilateral experiment between Germans, Israelis and Palestinians based on a trust game. By eliciting players’ beliefs over the
behavior of their counterparts we were able to compare beliefs with actual behavior. We found
substantial evidence for active trust and reciprocity even across borders, also reflected in beliefs
statements. While the levels of trust and reciprocity (beliefs) differ significantly in the three
countries with high Palestinian and low Israeli amounts we found no systematic discrimination
among participating subject pools.
Our most startling finding is the fact that players’ beliefs match well the actual behavior of
their counterparts from their own country but they misjudge the behavior of players in the other
countries. This observation has two important implications. Firstly, it reaffirms the differences
between countries in trust and reciprocity that we observed and suggests that these differences
prevail also outside the lab. Secondly, the fact that subjects’ beliefs attribute the standard of
behavior that prevail in their own environment also to subjects of the two other countries suggests
h
i
14
The rate of return is calculated as follows: R =
b̄
a
− 1 · 100 − 100.
16
interactions between individuals of different cultures to result with substantial surprises. The
most significant ones in our experiment are the interactions between Israelis and Palestinians.
Israeli senders make low transfers (to all subjects) and Israeli responders’ beliefs attribute low
transfers to Palestinians. In contrast, Palestinians senders make high transfers and Palestinian
responders expect Israeli senders to make high transfers as well. An interaction between an
Israeli sender and a Palestinian responders is expected to create a false perception of negative
discrimination, while an interaction between a Palestinian sender and an Israeli responders is
likely to result in an outcome that will be perceived as positive discrimination.
Our study on one-shot interactions revealed important and stable features of cross-border
interactions including the lack of intra-cultural and inter-cultural discrimination, stereotypes and
prejudices but the existence of different social norms and values across cultures. The discord
of actual behavior and beliefs in inter-cultural interactions can trigger and enforce distrusting
behavior, in particular if actions are interpreted as dicriminative that in fact are not but result
from different social norms. Constellations like in our study bear huge potential for conflicts in
repeated interactions. A key issue for future research is to develop a better understanding on
how conflicts arise, how they escalate and how they might be reduced in repeated scenarios.
References
Anderson, L. R., Y. V. Rodgers, and R. R. Rodriguez (2000), Cultural differences in
attitudes toward bargaining, Economics Letters 69(1), 45-54.
Bellemare, C., S. Kröger (2007), On representative social capital, European Economic
Review, 51(1), 183-202.
Ben-Ner A. and L. Putterman (2001), Trusting and trustworthiness, Boston University
Law Review, 81, 523-51.
Berg, J., J. Dickhaut, and K. McCabe (1995), Trust, reciprocity, and social history,
Games and Economic Behavior, 10(1), 122-142.
Boarini, R., J.-F. Laslier, and S. Robin (2002), Interpersonal comparison of utility in bargaining: Evidence from a transcontinental ultimatum game, Working Paper, Laboratoire d’Econométrie de l’Ecole Polytechnique, Paris.
Bond, M. H. and K. Hwang (1986), The psychology of the Chinese people, in M. H. Bond
(Ed.): The social psychology of the Chinese people, Hong Kong, Oxford University
Press 1986, S. 213-266.
Bornhorst, F., A. Ichino, K. Schlag, and E. Winter (2004), Trust and trustworthiness
among Europeans South - North Comparison, CEPR Discussion Paper No. 4378.
Bosman and van Winden (2002), Emotional Hazard in a Power-to-Take Game Experiment, The Economic Journal, 112, 146-69.
Bosman, R., H. Hennig-Schmidt and F. van Winden (2006), Exploring Group Decision
Making in a Power-to-Take Experiment, Experimental Economics, 9, 35 - 51.
Brandts, J., and G. Charness (2000), Hot vs. cold Sequential responses and preference
stability in experimental games, Experimental Economics 2(3), 227-238.
17
Brislin, R. W. (1970), Back-translation for cross-cultural research, Journal of Crosscultural Psychology 1, 185-216.
Buchan, N. R., Croson R. T. A., and Eric J. Johnson (2004a), When Do Fairness Beliefs
Influence Bargaining Behavior: Experimental Evidence from Japan and the United
States, Journal of Consumer Research, 31(2004): 181-190.
Buchan, N., Croson, N. T. A., and Johnson, E. J., (2004b): The boundaries of trust:
Own and other’s actions in the US and China. Journal of Economic Behavior and
Organization 55(4), 485-504.
Buchan, N. R., Croson R. T. A., and E. J. Johnson (2002), Trust and Reciprocity: An
International Experiment, Working Paper.
Cardenas, J. C., Carpenter, J., 2004. Experimental Development Economics: A review
of the Literature and Ideas for Future Research. Middlebury College Working Paper,
Middlebury College.
Castro, M. F., 2007. Where are you from? Cultural Differences in Public Good Experiments. Working paper, Department of Economics, Royal Holloway College, University of London, UK.
Cook, K. S. and R. M. Cooper (2003), Experimental studies of cooperation, trust, and
social exchange, in: E. Ostrom and J. Walker (Eds.), Trust and reciprocity, Russell
Sage, New York (2003), 209-44.
Chuah, S. H., Hoffmann, R., Jones, M., and G. Williams (2007), Do Cultures Clash?
Evidence from Cross-National Ultimatum Game Experiments, Journal of Economic
Behavior and Organization, forthcoming 2007.
Croson, R. T. A. Croson, and N. R. Buchan (1999), Gender and Culture: International
Experimental Evidence from Trust Games, The American Economic Review, 89(2):
386-391.
Danileson, A., and H. Holm (2005), Trust in the Tropics? Experimental evidence from
Tanzania, Economic Journal, 2005, 505-532.
Eco, U. (2000): Experiences in Translation, Toronto Italian Studies, University of
Toronto Press (2000).
Fischer, R. (2006), Congruence and Functions of Personal and Cultural Values: Do my
Values Reflect my Culture’s Values?, Personality and Social Psychology Bulletin, in
press.
Fukuyama, F., 1995. Trust: The social virtues and the creation of prosperity, The Free
Press, New York.
Fukuyama, F., 2001. Social capital, civil society and development. Third World Quarterly 22(1), 7-20.
Fershtman, C., and U. Gneezy. Discrimination in a Segmented Society: An Experimental
Approach, Quarterly Journal of Economics, February 2001, 351-377.
Goerg, S., Güth, W., Hennig-Schmidt, H., Walkowitz, G., and T. Weiland (2007), International Diversity of Fairness Concerns - An Online Ultimatum Experiment with
18
Germans, Israelis and Palestinians, Working Paper, Max Planck Institute for Research into Economic Systems, Jena, Strategic Interaction Group.
Greiner, B. (2004), An Online Recruitment System for Economic Experiments, in: V.
M. Kurt Kremer (Ed.), Forschung und Wissenschaftliches Rechnen 2003. GWDG
Bericht 63, Gesellschaft für Wissenschaftliche Datenverarbeitung, Göttingen, 79-93.
Güth, W., C. Schmidt and M. Sutter (2003), Fairness in the mail and opportunism in
the internet: A newspaper experiment on ultimatum bargaining, German Economic
Review 4(2), 243-265.
Henrich J. (2000), Does culture matter in economic behavior? Ultimatum game bargaining among the Machiguenga. American Economic Review, Vol. 90, No. 4,
973-979.
Henrich, J., R. Boyd, S. Bowles, C. Camerer, E. Fehr, H. Gintis, and R. McElreath
(2001), In Search of Homo Economicus: Behavioral Experiments in 15 Small-Scale
Societies, American Economic Review 91(2), 73-78.
Herrmann, B., C. Thöni, and S. Gächter (2007): Cooperation and punishment in a
cross-cultural perspective, Mimeo, University of Nottingham.
Hofstede, G. (2001): Culture’s consequences: Comparing values, behaviors, institutions
and organizations across nations, 2nd Edition, Sage Publications, Thousand Oaks.
Inglehart, R., M. Basañez, and A. Moreno (2001): Human Values and Beliefs, University
of Michigan Press, MI.
Koford, K. (2001), Trust and Reciprocity in Bulgaria: A Replication of Berg, Dickhaut
and McCabe. Working paper, University of Delaware.
Lazzarini, S., Regina, M., Rinaldo, A., and J. O. Siqueira (2004), Measuring Trust: An
Experiment in Brazil, Working Paper.
Luhmann, N. (1979), Trust: a mechanism for the reduction of social complexity, in: N.
Luhmann (Ed.), Trust and Power, Wiley, New York (1979), 4-103.
Parsons, T. (1957), The Social System, New York, Free Press.
Roth, A. E., V. Prasnikar, M. Okuno-Fujiwara, and S. Zamir (1991), Bargaining and
market behavior in Jerusalem, Ljubljana, Pittsburgh, and Tokyo An experimental
study, American Economic Review 81(5), 1068-1095.
Schwartz, S. H. (1992), Universals in the content and structure of values: Theoretical
advances and empirical tests in 20 countries. In M. P. Zanna (Ed.) Advances in
experimental social psychology (vol. 25, pp. 1-65). San Diego, CA: Academic Press.
Selten, R. (1967), Die Strategiemethode zur Erforschung des eingeschränkt rationalen
Verhaltens im Rahmen eines Oligopolexperimentes, in H. Sauermann (Ed.): Beiträge
zur experimentellen Wirtschaftsforschung, J.C.B. Mohr (Paul Siebeck), Tübingen,
136-168.
Walkowitz, G. (2004), Trust and Reciprocity in Transnational Interactions - An InterCultural Experimental Game with Germans, Argentineans and Chinese, University
of Bonn, mimeo.
19
Walkowitz G., Oberhammer C. and H. Hennig-Schmidt (2007), Experimenting Over a
Long Distance - A Method to Facilitate Inter-cultural Experiments, Working Paper,
Department of Economics, University of Bonn.
Willinger, M., C. Keser, C. Lohmann, and J.-C. Usunier (2003), A comparison of trust
and reciprocity between France and Germany Experimental investigation based on
the investment game, Journal of Economic Psychology 24(3), 447-466.
Appendix
A Instructions
1. Introduction
Thank you for participating in this international experiment. Please read the following instructions very
carefully. If you have any questions, please raise your hand. We will then answer your question in private at your
desk. During the experiment we will not answer any of your questions. In case you still have questions, please
read these instructions again. During the whole experiment it is very important that you do not communicate
with any of the other participants. We therefore ask you not to talk to each other.
To assure your anonymity and confidentiality you will have to randomly draw a code before the experiment
starts. When presenting this code you will later on receive your payments from this experiment. this procedure
ensures that we are not able to assign answers and decisions made in the experiment to specific persons.
In this experiment you can earn money. Your individual earnings depend on your own decisions and on the
decisions of other participants. You will get your personal payoffs on December XX to XX, from XX o’clock to
XXo’clock in room XXXX. Please watch over your code name and your code number carefully. You will need
them in order to collect your payment. Without code name and code number we cannot pay out your earnings.
Right after the experiment, you will receive a fixed fee of X US$ independent of the decisions you take in
today’s experiment. This fixed fee is paid in addition to the money you earned in the experiment.
During the experiment, you are involved into an interaction with a randomly assigned other participant.
This randomly assigned participant takes his decisions at a different point in time than you do without having
knowledge about your decisions. You are not told who this other participant is, neither during the experiment
nor at any point in time after the experiment. The persons who participate with you in the experiment today are
not assigned to you.
After this introduction, we will ask you some questions to check your understanding of the experimental
rules. Then the experiment itself will start. After the experiment, we will ask you to fill in a questionnaire we
need for a statistical analysis of the experimental data.
2. The general design of the experiment
During the experiment, pairs of two participants are formed who interact. These participants are called person
A and person B from now on. All persons in one room are either persons A or persons B. Thus, persons A and
persons B cannot be in one room at the same time. All persons who participate with you in the experiment at
this moment belong to the same type of person.
The experiment consists of three decision rounds structured identically. Within these rounds, each participant
takes decisions keeping the same role, either as person A or as person B.
At the beginning of each round, each person A and each person B receives an initial endowment X of 10 ECU
(Experimental Currency Unit) the exchange rate being 1 ECU = 0.50 US$. Person B can keep this money. This
amount will be paid to her at the end of the experiment, together with the payment of her experimental earnings.
Person A can now choose to either send an integer part, all or nothing of his initial endowment X to person
B. This means, he can choose any integer a from the interval [0,10] of his initial endowment he wants to send to
20
person B. Each amount a chosen by person A is tripled. Thus, person B does not receive the amount a originally
sent by person A but she receives three times as much, i.e. 3a.
Here is an example: If person A sends 4 ECU, person B receives 12 ECU. If person A sends 6 ECU, person B
receives 18 ECU.
During the experiment, Person B is not told which amount a person A actually chose. Therefore, person B
does not know how large 3a actually is. Person B has to decide how much she wants to transfer back to person
A. Since person B does not know the actual amount a or 3a respectively having been sent by person A she has
to decide for each possible value of the tripled amount 3a how much she wants to transfer back to person A. The
amount person B transfers back to person A is b. Person B can decide whether she wants to transfer back an
integer part, all or nothing of 3a to person A. The amount b transferred back is not multiplied. Person A receives
no information about person B’s actual decisions during the experiment.
Person B’s decision table for one round looks as follows:
Person A sent you a:
You receive 3a:
0
0
1
3
2
6
3
9
4
12
5
15
6
18
7
21
8
24
9
27
10
30
Which amount b do you
want to transfer back?
After B’s decision on b, the first decision round has finished for both persons. The general procedure of the
following second and third decision rounds corresponds to the procedure of the first decision round.
Actual payments of the experimental earnings of person A and person B are made after all experimental
sessions having been finished. By assigning person A’s actual decision a to person B’s decision b being valid for
person A’s actual decision a, individual payment for each round can be calculated.
Person A’s individual profit GA for one round is:
Initial endowment X
- amount a sent
+ amount b transferred back by person B
= GA , Person A’s individual profit for one round
Person B’s individual profit GB for one round is:
Initial endowment X
+ tripled amount a, sent by person A
- amount b transferred back by person B to person A
= GB , Person B’s individual profit for one round
21
Here is an example: Person A decides to send 4 ECU to person B in one round. Then, Person B receives 12
ECU. Person B decides to transfer 5 ECU back.
Person A’s individual profit GA for this round is:
Initial endowment X
10 ECU
- amount a sent
-4 ECU
+ amount b transferred back by person B
= GA , Person A’s individual profit for this round
+5 ECU
=11 ECU
Person B’s individual profit GB for this round is:
Initial endowment X
10 ECU
+ tripled amount a, sent by person A
- amount b transferred back by person B to person A
= GB , Person B’s individual profit for this round
+12 ECU
-5 ECU
=17 ECU
The individual total experimental payoff for each participant is the sum of payoffs in all three rounds plus the
show up fee of X US$ you will be paid today.
3. The experimental procedure
In the following, we precisely explain how the experiment proceeds and which consecutive decisions you will have
to take.
Please read this instruction very carefully. We then will answer your questions in private. Please raise your
hand if you have questions. Before the experiment starts, we again will ask all participants whether they fully
understood the instructions of the experiment.
We then will give you a short exercise to check your understanding of the experimental rules. After having
solved these exercises, the experiment will start.
Start of the experiment
At the beginning, each participant randomly draws a code name. You will need the code name for being paid
your experimental earnings. Furthermore, it guarantees full anonymity. Please watch over this code name very
carefully.
Then, each participant randomly draws a decision sheet for each decision round. Each decision sheet contains
your personal code and information about persons A or B assigned to you. Please note that the assignments as
well as the order of assignments of the persons you are interacting with are randomized. On the slip containing
your code you find an empty space. Please fill in your code name there.
Decision procedure for person A:
Round 1
You learn that you are person A.
Person A is paid an initial endowment of X=10 ECU by the experimenters.
Person A learns that a person B was randomly assigned to him by the experimenters. Person A learns the
name of the university, the city, and the country where the assigned person B is studying. With this participant,
person A is interacting during this round.
Person A decides which amount a he actually wants to sent to the randomly assigned participant B. He can
chose any integer from the interval [0,10].
After having decided on a, the first decision round for person A has finished.
Round 2
22
The procedure of this decision round for person A is identical to that of round 1. However, in this round a different
interacting person B is randomly assigned to person A. This means that in this second round person A does not
interact with the same person B from round 1.
Round 3
The procedure of decision round 3 for person A is identical to the procedure of the previous rounds 1 and 2.
However, again a different interacting person is randomly assigned to person A. This means that in this third
round, person A does not interact with one of the persons B in rounds 1 or 2. During the whole experiment, each
person interacts with an assigned person B only once.
The experiment has now finished for person A.
Decision procedure for person B:
Round 1
You learn that you are person B.
Person B is paid an initial endowment of X=10 ECU by the experimenters. Person B owns this amount and
keeps it. She will be paid this amount later, together with her experimental earnings.
Person B is told that a person A was randomly assigned to her by the experimenters. Person B learns the
name of the university, the city, and the country where the assigned person A is studying. With this participant,
person B is interacting during this round.
Person B is asked which amount she wants to transfer back to person A for each possible value of 3a. Person
B indicates an integer for all 11 possible amounts of 3a being the amount b person B wants to transfer back to
person A.
After having decided on b, the first decision round for person B has finished.
Round 2
The procedure of this decision round for person B is identical to the procedure of round 1. However, in this round
a different interacting person A is randomly assigned to person B. This means that in this second round person
B does not interact with the same person A from round 1.
Round 3
The procedure of decision round 3 for person B is identical to the procedure of the previous rounds 1 and 2.
However, again a different interacting person is randomly assigned to person B. This means that in this third
round, person B does not interact with one of the persons A in rounds 1 or 2. During the whole experiment, each
person interacts with an assigned person A only once.
The experiment has now finished for person B.
At the end of the experiment we would like you to answer some questions we need for the statistical analysis of
this experiment.
23