The influence of social distance and strategic consideration on reciprocity
– the evidence from the investment game experiments
Author: Wojciech Czajkowski
Supervisor: Dr. Silvia Dominguez Martinez
Master thesis, MSc Business Economics, University of Amsterdam
Abstract
The research provides insight into changes of reciprocal behaviour under two different conditions: (1) an increased social distance, interpreted as the indirect relation between parties, and (2) a strategic concern in providing kindness. 270 high school and university students took part in the series of the economic experiments, based on a modified version of Berg et. al. (2005) investment game. The results show that (1) participants are on average less willing to reciprocate kindness if the relation between giver and receiver is less direct and that (2) consideration of strategic goal in kind behaviour reduces reciprocal behaviour of kindness receivers.
Keywords: Reciprocity, Investment game, Strategic consideration, Social distance
Statement of originality
This document is written by Student Wojciech Czajkowski who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of contents
Abstract ... 1
1. Introduction ... 3
2. Literature review ... 4
2.1. Fairness and reciprocity theory ... 4
2.2. Investment game experiments... 6
2.3. Modifications to the investment game ... 7
2.4. Contribution to the current literature ... 8
2.4.1. Indirect relation ... 8 2.4.2. Strategic consideration ... 9 3. Methodology ... 9 3.1. Experimental design... 9 3.1.1. Indirect relation ... 11 3.1.2. Strategic consideration ... 12 3.2. Experimental procedures ... 13 3.2.1. Strategy method ... 13 3.2.2. Currency ... 14 3.3. Theoretical predictions... 14 3.3.1. Indirect relation ... 14 3.3.2. Strategic goal ... 15 4. Results ... 16 4.1. Experimental setting ... 16 4.2. General results ... 17
4.2.1. Normality and use of statistical methods ... 18
4.2.2. Distribution among samples ... 19
4.3. Social distance ... 20
4.4. Strategic consideration ... 22
4.4.1. BONUS treatment ... 22
4.4.2. DICT treatment ... 24
4.5. Additional analysis – regressions... 24
5. Discussion ... 27
5.1. Limitations ... 27
5.2. Conclusions and further research ... 29
6. References ... 31
Appendix A – questionnaire ... 34
1. Introduction
The economic concept of selfish individuals shaped a wide range of important economic theories. However, even classics, proposing the principle of self-interest as the basis of economics, were aware that such a principle is not actually realistic (Edgeworth, 1881; Sen, 1977). The conceptual absolute selfishness has been therefore superseded by theories of fairness preferences and conditional behaviour of individuals. The aim of this research is to explore the nature of reciprocity – the phenomenon receiving due attention by many recent economics researchers. The experimentally proven importance of reciprocity in economics is unquestionable, from the role of shaping the interactions between individuals to providing efficiency in the markets based on incomplete contracts.
The present research aims to answer two questions concerning the reciprocal behaviour of individuals. Firstly, it tests how people react in case of indirect relation (larger social distance) between providers and recipients of kind behaviour. Secondly, the paper examines whether reciprocity of receivers of kindness changes when the intentions of giving may be considered as selfish (strategic). To answer these problems, a series of economic experiments was organised, during which 270 high school and university students took part in the investment game (Berg, Dickhaut, & McCabe, 1995), specially modified for the purpose of this research. The experimental findings showed that receivers reciprocate kindness to a lesser degree in larger social distance treatment, in which the kind decision of sender affected multiple receivers, as a contrast to the standard one-on-one game. Moreover, the strategic consideration was found to affect the reciprocal behaviour of players. When the monetary motivation for providers to send funds, absent in the first round, was introduced in the second round, almost two-thirds of receivers significantly decreased the amounts sent back. In the treatment where the strategic goal in sending funds was strengthened in the second round (by additional bonus), the majority of receivers sent back less in round two than in round one. However, in this treatment, a substantial number of players increased the amounts sent back in the round with a bonus in order to help senders earn the additional funds.
The inspiration for choosing the examined features influencing reciprocity came from market observation. Well covered by literature is the fact that reciprocity ensures the market efficiency when designing and executing complete contracts is unrealistic. However, not only incomplete contracts benefit from the existence of reciprocity. Marketing strategies, on which companies spend hundreds of billions of dollars worldwide each year, often rely on direct and indirect reciprocity of targeted clients. Many firms provide diverse kinds of gifts to (potential) clients as an integral part of their marketing strategy. In the survey organised for the purpose of this
experiment1, 48% of 205 respondents claimed that they received some kind of gift from some firm within one year prior to the survey, which shows a large scale of gift providing by companies. In the same survey, 63% of respondents indicated that they feel less obliged to reciprocate the kindness towards firms (for example, by buying something) than towards any individual, while only one-fourth indicated that they are as reciprocal towards firms as towards individuals. Asked for the reason of a different approach to firms, the majority indicated a different purpose of providing gifts by firms and a less close relation between firms and clients. Considering the core questions of this research, 71% of the surveyed sample claimed that in their opinion firms give presents strategically, in order to gain monetary profits on clients. 73% of respondents reported that they feel treated as a group of clients when receiving a present from a firm, therefore not feeling close relation. An interesting issue seemed to be to check experimentally whether features perceived as characterising a client-firm relation indeed influence the reciprocal decisions of people. It should be noted that the behaviour studied in this research generalise onto many more contexts than only the afore-mentioned relation between firms and clients.
The thesis is organised as follows: chapter two contains a review of the relevant literature, chapter three describes the design and procedures of experiments, together with theoretical predictions. In chapter four the results of experiments are presented and analysed. Finally, in chapter five the limitations and conclusions are discussed.
2. Literature review
In this section, the main concepts of fairness and reciprocity are briefly outlined. The game on which the experiment bases is introduced, with examples of its use in various studies. Furthermore, it is explained how the present study expands the available research on the topic.
2.1.Fairness and reciprocity theory
The classical economic principles of a selfish, profit-maximising individual are ousted by the influence of more recent studies concerning behavioural aspects. The number of theories describe regular deviations from the selfish axiom, one of which is the existence of reciprocity in people’s behaviour.
The reciprocity is defined in a twofold manner: as a cooperative and more friendly response to a niceness than the self-interested theory would predict and as a retaliatory and more hostile
reaction to an unfriendly action to the contrary (Fehr & Gächter, 2000). The former is an example of a positive reciprocity, while the latter describes the negative reciprocity. Reciprocity occurs when the described reactions are observed even if no material gain or a personal cost is expected as a consequence. Moreover, reciprocity has to be conditional on some action of others – the unconditional niceness is defined as altruism.
Considering the behavioural theories of fairness and reciprocity, outcome-based and intention-based approaches can be distinguished. The former type can be represented by the model proposed by Fehr & Schmidt (1999), which states that the utility of a person depends on their own given level of inequity aversion. In the model of two players, Player 𝑖 can have a certain disutility from inequitable payoff between him and Player 𝑗. This disutility can exist both in case the difference favours Player 𝑗 (the second term in the function below with coefficient 𝛼𝑖) and in case the difference is advantageous to Player 𝑗 (the third term with coefficient 0 ≤ 𝛽𝑖 < 1). Moreover, the authors assume that subjects are loss averse and suffer more from inequity with their material disadvantage than from advantageous inequity (𝛼 ≥ 𝛽) , since in the opposite case a person would increase utility by throwing out money. The whole utility function of Player 𝑖 in the two-player case is presented below.
𝑈𝑖(𝑥) = 𝑥𝑖 − 𝛼𝑖 𝑚𝑎𝑥{𝑥𝑗− 𝑥𝑖, 0} − 𝛽𝑖 𝑚𝑎𝑥{𝑥𝑖− 𝑥𝑗, 0}, 𝑖 ≠ 𝑗.
The model proposed by Fehr and Schmidt assumes that people are concerned with outcomes of actions but neglects the importance of others’ intentions on certain players’ decisions. This question is raised by the intention-oriented theories. Rabin (1993) defines fairness equilibrium using a game theoretical framework and emphasises the role of reciprocity in players’ choices. The following theory of Sequential Reciprocity (Dufwenberg & Kirchsteiger, 2004) puts weight on the beliefs aspect, already introduced by Rabin. Dufwenberg & Kirchsteiger claim that since reciprocity is dependent on intentions, and intentions are caused by beliefs, then reciprocal motives are based on “believes about believes”. The authors extend Rabin’s work from normal form games to extensive form games, forming and applying the sequential
reciprocity equilibrium.
The two above-mentioned approaches to fairness are compared and tested empirically by McCabe, Rigdon, & Smith (2003). In the voluntary trust game (VTG) and involuntary trust game (ITG) used by the authors, the outcome-based Fehr and Schmidt model predicts the same level of cooperation across the games, regardless of whether one of the players is forced to cooperate or not. Due to the intention-based approach, the inability to signal intentions by one of the players should reduce the amount of cooperation in ITG (formed as a Trust and Reciprocity hypothesis). The experimental results show a significant difference between VTG
and ITG results, which was predicted by intention-based Trust and Reciprocity hypothesis, strengthening the advantage of intention-based concepts and the role of reciprocity over stable fairness preferences.
2.2.Investment game experiments
To study the effects of reciprocity and trust, Berg, Dickhaut, & McCabe (1995) designed the double-blind, one-shot game of two-players exchange. The general rules of a game are that one of the players (sender) sends some amount to the receiver, then the amount is tripled by the time it reaches the receiver and next the receiver can send some of it back to the sender. The detailed description of the original investment game is included in chapter 3 – Methodology. The results of the experiments in Berg et al. (1995) proved that receivers show reciprocal behaviour by lowering their own payoffs in favour of the payoff of the other player, and that the senders anticipate such behaviour and trust in the first move. The design of the game allowed to reject the possibility of self-interest driving the behaviour of players alone. Results also consequently contradicted the subgame perfect Nash equilibrium.
The investment game (Berg et al., 1995) has been widely used by numerous researches investigating the reciprocity and trust concerns. A wide body of researchers seeks for the difference in behaviour in certain groups of subjects. Buchan, Croson, & Solnick (2008) used it to find that on average men are more trustful but less trustworthy (positively reciprocal) than women. Also, their attitude is more strategic since the expected return on trusting for men was significantly higher than for women. The difference between genders’ behaviour was high in case of senders, but the difference in case of receivers was relatively small (four percentage points). Both genders were trusted equally and perceived as equally trustworthy; also the differences in interaction between genders turned out to be insignificant. Similar research done by Chaudhuri & Gangadharan (2003) found a significant difference between genders in money sent (trust), but in case of money returned (reciprocity) the difference was not significant in most statistical tests.
Willinger, Keser, Lohmann, & Usunier (2003) compared the investment game results between people from different countries – Germany and France – and found a significantly higher level of trust (senders) in Germany but no significant difference in the reciprocal behaviour (receivers) between the nations. These results were also compared to the original (Berg et al., 1995) results from the USA and Germany turned out to have significantly higher levels of trust and reciprocity than the USA, while the differences between France and the USA were not statistically significant. The results of the investment game from other two countries – Austria and Japan – were compared by Akai & Netzer (2012). Their findings showed identical
reciprocity and trust levels in the two countries when subjects were playing with counterparts from the same nation, but significant differences between countries in case of international reciprocity and trust.
2.3.Modifications to the investment game
Also, the robustness of the investment game design has been examined by various studies. The question whether the decisions in investment game are dictated by risk attitudes of the players was asked in the research of Houser, Schunk, & Winter (2010). In their experiments, players’ attitude towards risk predicted the behaviour in the modified investment game where the receiver’s role was played by the computer, but the trusting decisions in an actual investment game between two people were not predicted by risk attitudes. The interpretation of these results was that there is no strong connection between risk attitudes and the trust and reciprocity decisions.
Charness, Cobo-Reyes, & Jiménez (2008) added a third person to the original investment game to test for a shift in fair behaviour under the perspective of being monitored and punished by the additional player. The third party was not affected financially by the other players’ decisions but could punish selfish receivers or reward senders who were harmed by the decisions of the former. They found that inclusion of such third party significantly increased the amounts sent by both senders and receivers under either scenario: punishment and reward. The indirect reciprocity was studied in another third-party inclusion in the investment game, done by Güth, Königstein, & Marchand (2000). In their indirect treatment, the sender was not rewarded by the receiver to whom he transferred the money, but by the receiver from another pair who observed the sender’s choices. The outcome showed positive amounts invested and sent back by players in indirect treatment, proving the existence of indirect reciprocity; however, these amounts were significantly lower than in the standard investment game. Coricelli, Morales, & Mahlstedt (2006) introduced asymmetry in the information available in the investment game. In their modified design, there were two possible multipliers of the amounts sent to the receiver. Only the receiver learned the multiplier so that in case the sender received back a low amount back, the sender did not know whether it had been caused by the receiver’s selfishness or a low multiplier drawn. Under this asymmetric information conditions the reciprocal behaviour of receivers, as well as the trust of senders, remained the same as in the previous experimental studies with complete information.
A vast number of 162 studies based on Berg et al. (1995) investment game were collected in a meta-analysis done by Johnson & Mislin (2011). The features shown to be influential on the
players’ decisions in analysed experiments were some of the geographic regions, the rate of return on trust (the amount multiplier), using students as the subjects, playing against a computer and one subject playing both roles. An interesting fact is that the use of random pay, perceived by many studies as not driving the results, turned out to cause a high and significant decrease of the amount sent, whilst having no influence on reciprocity. Moreover, the use of a double-blind protocol had no significant impact on the average player’s decisions. The anonymity of players gave a statistically weak result of decreasing trust and increasing reciprocity in some regression specifications.
2.4. Contribution to the current literature 2.4.1. Indirect relation
To my best knowledge, the studies focusing on reciprocity in a situation of larger social distance2 meant by the indirect connection between participants is highly limited, and I was unable to find studies conducting any comparable experimental research in this area. The effect of a different number of players in economic games was examined in Isaac & Walker (1988) research about group effects within voluntary contribution mechanisms3, as well as in the following studies based on VCM. In this experiment, however, each of the objects of interest – receivers – is playing with one sender as in the original investment game, and each receiver’s decision is applied to only one person. This research is focused on a less direct connection instead of a larger amount of players to which one’s decision may apply. In this way, the study may resemble the experiments done by Dana, Weber, & Kuang (2007), which showed that a weakening connection between players’ choices and their effects on others’ payoff can significantly reduce people’s pro-social behaviour in the modified dictator game. In contrast to this research, the decisions in Dana et al. (2007) experiments did not have a guaranteed result on others’ payoff and the effect of such uncertainty played a big role in interpreting the results. Moreover, the results of (Dana et al., 2007) and alike dictator games experiments showing the tendency to shirk once players were given various excuses to do so, have been questioned by the work of Weele, Kulisa, Kösfeld, & Friebel (2014). The authors undermine the generalizability of previous dictator game works to different scenarios. They used the trust
2 Social distance in most studies is explained by an expected level of reciprocity by subjects
within an interaction (Hoffman et al., 1996) and usually induced by greater anonymity (Cox & Deck, 2005; Hoffman et al., 1996)
3 Isaac & Walker (1988) find that the contribution to public good decreases mostly due to
game and the moonlighting game, both similar in design to the investment game, and obtained drastically different results than dictator game-based studies: the frequency of the players’ reciprocal behaviour did not change under the existence of the excuse for unfair behaviour.
2.4.2. Strategic consideration
Similarily, the effect of the existence of a strategic consideration on the reciprocal behaviour of (kindness) receiver has not been covered by the existing literature so far. The studies using the investment or trust games did not assess how intentions of profit-maximisation influenced the reciprocating player's behaviour. The intention-based reciprocity theory suggests that such difference can have a big impact on the decision process of players. In the roots of these theories lies the conviction that intentions form reciprocal behaviour (Rabin, 1993). Thus examining the influence of shifting the perceived intentions on the decisions may help to assess the strength of the connection between intentions and reciprocity. The experiments based on the trust game showed that the inability to read the first-mover’s intentions reduces the likelihood of observing cooperation (McCabe et al., 2002), but the field lacks studies investigating how the different intentions affect the reciprocity.
3. Methodology
In this section, the Berg et al. (2005) investment game is explained and the modified design of this particular experiment is described. Next, the experimental procedures are introduced. Finally, the theoretical predictions are discussed.
3.1.Experimental design
The game used in the experiment is based on the earlier mentioned investment game (Berg et al., 1995). The original game is set between two types of players: sender (player A) and receiver (player B), sitting in two separate rooms. Both player types receive $10 as a show-up fee. While each player B saves the amount, each player A can decide to send some of their show-up fee (𝑀) to an anonymous player B. Player A can send from zero to all of the show-up fee:
$0 ≤ 𝑀 ≤ $10
Every dollar sent is tripled (3𝑀) by the time it reaches the room with B players, which is common knowledge. Players B then decide (𝑘) how much of the received amount to send back to player A (𝑘(3𝑀)):
This results in a payoff for player A of 𝑃𝐴 and a payoff for player B of 𝑃𝐵 shown below: 𝑃𝐴(𝑀, 𝑘) = $10 − 𝑀 + 𝑘(3𝑀)
𝑃𝐵(𝑀, 𝑘) = $10 + 3𝑀 − 𝑘(3𝑀)
The authors ran two treatments, a no-history and a history treatment. In the latter, each counterpart received a report summarizing the decisions and results of all pairs of players from previous no-history treatment. Given that in this particular study the subjects do not receive feedback, only the no-history treatment is considered.
Using backward induction to solve the investment game, the dominant strategy of players B is to send back $0 for all amounts received. Anticipating this, the dominant strategy of players A is also to send $0. Therefore, the unique Nash equilibrium of the Berg et al. investment game is to send no money by both types of players.
In this thesis, an adjusted version of Berg et al. (2005) investment game is used. The experiment consists of four treatments, with two rounds per treatment, as shown in Figure I. 1-1 and 1-4 treatments have the purpose of testing the differences in behaviour caused by larger social distance. In case of these treatments the between sample variation is used – the subjects take part in standard 1-1 or modified 1-4 session. The BONUS and DICT treatments are used to examine the effect of strategic consideration on reciprocal behaviour, BONUS by strengthening the strategic factor in the second round, and DICT by taking the strategic factor away in the first round and then including it in the second round. In these treatments the within-sample variation is used – the subjects take part in a baseline round and a treatment round (BONUS or DICT).
FIGURE I
Four types of treatments scheme 1-1 BONUS treatment
1st round: Baseline round 2nd round: Bonus round
1-4 BONUS treatment 1st round: Baseline round 2nd round: Bonus round
1-1 DICT treatment 1st round: Quasi-dictator round 2nd round: Baseline round
1-4 DICT treatment 1st round: Quasi-dictator round 2nd round: Baseline round
3.1.1. Indirect relation
The variation between the 1-1 and 1-4 treatments is designed to check for the influence of a less direct relation between the players on their behaviour. The first treatment type, called 1-1, is an investment game set between two individuals to measure the behaviour exactly like in the original version of the game. The new 1-4 treatment type makes one player A’s decision applied to four randomly chosen players B, as shown in Figure II below.
FIGURE II 1-4 treatment scheme
The amount 𝑀 chosen by each player A is tripled (3𝑀) and sent to four randomly chosen B players, so that each of these B players receive the same amount. Next, each of these players Bi individually chooses how much money to return. Player A receives then the average of the four B players’ contributions so that the fact of playing with multiple receivers does not induce monetary advantage for the sender. This setup is chosen to measure the effect of relaxing social connections without allowing for the possible payoff inequity effects. The payoff of player A (𝑃𝐴) and of each player B (𝑃𝐵𝑖) in 1-4 treatment are as follows:
𝑃𝐴 = $10 − 𝑀 +
1
4∑𝑘𝑖(3𝑀), 𝑖 ∈{1,2,3,4} 𝑃𝐵𝑖 = $10 + 3𝑀 − 𝑘𝑖(3𝑀)
The 1-4 treatment creates a less direct relation between players A and B than 1-1. The payoff of player A no longer solely depends on the decision of one player B like in the original game, but relies on four players’ B decisions, implying larger social distance.
Player A 𝑀
𝑘4(3𝑀)
Player B2
3𝑀 3𝑀 3𝑀 3𝑀
Player B1 Player B3 Player B4
𝑘1(3𝑀) 𝑘2(3𝑀) 𝑘3(3𝑀)
Player A: 1
3.1.2. Strategic consideration
The difference between rounds in each 1-1 and 1-4 session manipulates with the strategic consideration in sending funds to the receiver. Recall that Figure 1 presents the scheme of all experimental treatments. There are two types of treatments designed to tease out the effect of strategic consideration on reciprocity: BONUS and DICT, both in 1-1 and 1-4 versions. In the BONUS treatment the first round is a standard (baseline) investment game, and in the second round, the strategic goal is strengthened by including an additional motivation for player A to send money.
In the second round of BONUS treatment, player A can get $10 bonus if the average of the players’ B contributions exceeds some minimum value (threshold, 𝑇). In this way the payoff of player A equals:
In 1-1 treatment: 𝑃𝐴 = $10 − 𝑀 + 𝑘(3𝑀) + 𝛼($10), where 𝛼 = 1 iff 𝑘(3𝑀) ≥ 𝑇, otherwise 𝛼 = 0 In 1-4 treatment: 𝑃𝐴 = $10 − 𝑀 + 1
4σ 𝑘𝑖(3𝑀) + 𝛼($10), where 𝛼 = 1 iff 1
4σ 𝑘𝑖(3𝑀) ≥ 𝑇, otherwise 𝛼 = 0
Both types of players are aware of the value of the bonus and the existence of a threshold, but only players A know the specific threshold value. In this way, players B are unable to choose a response so that payoffs would be equal. Such a solution is chosen to emphasise to players B that players A have the additional incentive to send money, instead of pushing players B to calculate the exact player’s A payoff.
One may argue that strategic consideration exists in the investment game even without being strengthened with a bonus. To search for a more pure effect, an additional DICT treatment type with a different approach has been implemented. In DICT treatment, the strategic goal in sending money by the player A is absent in round one and then introduced in the second, baseline round.
The first round of the DICT treatment is divided into two parts. In the first part, after the players are assigned a role, everyone is informed that player A can send some amount to player B and that the sent amount will be tripled. Then the answer sheets are distributed to players A, while players B are given a card with the information that they do not take any decisions in this part of the experiment. The cards are distributed to both types so that it is harder to tell out what role are the surrounding participants assigned to.
After players A make their decisions, the cards are collected and the second part of instructions is distributed which informs the players about the possibility for players B to send some of the received amount back to player A. Then the answer sheets are distributed like in the first part of round one, and after the decisions are made by players B the first round ends. The second round is a standard round with complete information. By using such design, in the first round of the DICT treatment players A make decisions similar to the ones made in the dictator game (hence the name of the treatment). Players B know that during the decision making, players A were unaware of the possibility of receiving any amount back, so that their decisions were not strategic. In the second round, players have complete information, just like in the standard investment game, therefore players B know that players A could count on receiving some (multiplied) amount back.
3.2.Experimental procedures
The course of the experiment is as follows: participants are given general instructions and first-round instructions with numbers of players written on them. After the instructions are read out loud by the experimenter, the answers to the control questions are checked to examine the understanding of the rules. After that, the answer cards are provided and collected. Then in case of BONUS treatment the round ends, whereas in two-step DICT round the second part of round one starts. Round two is organised analogously to the first one. After completing the second round, the questionnaires are distributed, and during their completion, the winners are drawn by the experimenter and the local tutor, using the players’ numbers, and the winners’ payoffs are prepared. After completing the questionnaires, when leaving the classroom, the selected players collect payoffs placed in envelopes, next to their player numbers and receipts to sign. In this way, the conditions as close to the double-blind procedure as possible are created.
3.2.1. Strategy method
To gather players’ B choices, the strategy method is used. It means that both types of players make decisions simultaneously, not knowing the decision of the others. Since players B are responding to players’ A choice, they choose what amount to give back for every possible amount received, so in total players B make 10 decisions. The effects of the use of the strategy method in experiments are discussed by Brandts & Charness (2011), who find that the impact of its use is limited, and in no case the effect of treatment found using strategy method was not observed with the direct-response method. Moreover, there is no reason to expect any potential effect of the strategy method to be different between treatments, which could eventually bias the results.
3.2.2. Currency
The experiments are arranged in Poland, therefore the currency of the game is PLN (zloty, abbreviation: zł). The amounts in the game are exchanged in a 1-1 ratio since the predicted payoffs are close to the usual payoffs from the economic experiments in Poland4. In each session, at least one pair of players is randomly selected and the payoff from one random round is paid to them. Such payoff perspective was assessed as sufficiently incentivising by economics teachers and lecturers of high school and university where the experiments took place.
3.3.Theoretical predictions
Both examined factors – indirect relation between giver and receiver and the strategic concern in giving – are widely encountered in social relations and due to economic theories, both can have a significant impact on people’s behaviour. Basing on theory and experimental evidence, some predictions about players’ behaviours are formulated into research hypotheses.
3.3.1. Indirect relation
The relaxation of the one-on-one connection between parties can lead gift receivers to exploit the moral excuse not to reciprocate, due to the fact that their response does not determine player A’s payoff completely. The receivers may also perceive the social distance between them and player A to be larger because the decision of player A is not made directly towards them, but towards the random four people. This can in turn make players B feel less obliged to reciprocate the favour, compared to the situation if the relation was direct. Experiments using the slightly modified trust game found that the higher social distance weakens the effect of positive reciprocity (Cox & Deck, 2005). The same results have been obtained using the dictator game (Hoffman, Mccabe, & Smith, 1996)5. Modification very similar to the 1-4 treatment was introduced in the research of Maximiano, Sloof, & Sonnemans (2004), who studied the gift exchange game in one employer and four employees setting. In their design, multiplying employees (receivers) went together with multiplying the amount received back, so that the
4 The average payoff for participation in an economic experiment in University of Warsaw is
around 40zł for a 75-90min session. The possible payoffs in the present experiments are from 0zł to 40zł for a 40min session, depending on a role played. The average payoffs based on players’ choices are calculated in the Results chapter.
5 As the factor changing the social distance, in both cases authors used the single blind versus
sender was, in fact, playing four gift exchange games at once. Thus the monetary advantage gained by the senders made it hardly possible to measure the pure effect of the larger social distance. Moreover, the framing of players as employees and workers could have also influenced the results, encouraging employees (receivers) to return an effort for salary as in everyday work relation, thus suppressing the negative impact of the treatment on reciprocity. Nonetheless, Maximiano et al. (2004) found slightly lower effort levels provided by employees in 1-4 treatment compared to 1-1. Taking into consideration the aforementioned researches and the theory, the first hypothesis is:
H1: People are less willing to reciprocate kindness if the relation is indirect (the social distance is larger)
The hypothesis will be tested by comparing of the results between of 1-1 and 1-4 versions of consecutive rounds, separately for DICT and BONUS treatments. According to the hypothesis, the amounts sent back by the receivers in 1-1 treatments, both in the first and the second rounds, are expected to be higher than in 1-4 treatments, where the effects of players’ B decisions on players’ A payoff are weaker.
3.3.2. Strategic goal
In case of the existence of a strategic goal in providing kindness (such as sending money), the intentions of the sender can be read differently by the receivers than without such goal. If the gift receiver realises the strategic (selfish) aim of the gift provider, the act of giving can be no longer perceived as an act of niceness or trust, but as a calculated attempt to maximise the profit. Moreover, if the receiver interprets the intention of the gift provider as selfish, the negative reciprocity may encourage the receiver to act selfishly as well and reverse the effect of positive reciprocity for the gift the provider would count on. Consequently, strengthening the strategic goal in sending amounts to the receiver can cause a decrease of the receiver’s positive reciprocity or strengthen the negative reciprocity if selfish motives were already suspected. Numerous experimental results6 find support for the intention-based models of reciprocity and inconsistencies of outcome-based models. This fact can indicate that motives can be almost as important as the gift itself. The intention-based approach, suggesting the possible change of
6 Described in the Literature review section (e.g. McCabe et al., 2003; Coricelli et al., 2006;
the individual’s behaviour caused by the change of intentions, leads to the following second hypothesis:
H2: The strategic consideration in providing kindness decreases reciprocity of the kindness receiver
To test the H2 hypothesis, the results between rounds within each treatment will be compared. In line with the hypothesis, the amounts sent back by players B should be higher in the first round of DICT treatment compared to the second round, since the first round lacks a strategic goal in providing a positive amount by players A. By analogy, the amounts sent back by players B in the control round in BONUS treatment should be higher than in the second round, where the goal in sending positive amounts to the receivers is strengthened. The differences between rounds in each treatment will be tested for both 1-1 and 1-4 sessions.
4. Results
In this chapter, the results of the experiments are presented. First, the general aspects such as the setting of the experiments, descriptive statistics, differences between the samples and the choice of statistical methods are discussed. Then the treatments outcomes are compared and analysed to test the predictions of the research hypotheses. In the last part, the regressions are included to check whether other observables affect the players’ reciprocal behaviour, in a way that could undermine the causal effects of treatments.
4.1.Experimental setting
Three series of economic experiments have been run in total. The first series, organised in First High School in Świdnik, Poland, included 150 second and third-year students taking part in seven experimental sessions – four 1-1 and three 1-4 versions, all of which were the BONUS treatments. The second series of experiments took place at UMCS university of Lublin, Poland. 120 students of the third year of economics took part in seven experimental sessions, four of which were BONUS treatments (two 1 and two 4) and three were DICT treatments (one 1-1 and two 1-1-4 treatments). With only a few exceptions, neither high school nor university students had any previous experience of participating in a similar economic game. The general descriptive statistics are shown in Table I.
TABLE I
Descriptive statistics divided by two experimental samples
Sample Number of all players Players B (male/female, %)Gender share Age
High School 150 98 56/44 18-19
University 120 80 46/54 22
Total 270 178 49/51 20
4.2.General results
The overall understanding of the game procedures, self-reported in questionnaires after the experiments, was highly satisfactory, with 92% of all participants claiming they understood the rules clearly. On average, subjects from both samples – high school and university students – showed high levels of trust and reciprocity. Of 98 players B in high school, only 5 sent back zero for every possible amount in both rounds; in case of university students, 3 of 80 players B consequently sent back nothing in both rounds. Similarly, no player A in high school or at the university sent zero in both rounds to player B. Comparing these results to original Berg et al. (1995) findings, fewer of the players A consequently chose selfish behaviour (2 of 32 players A in the original game). A comparison of players’ B behaviour is less straightforward due to the strategy method used. Berg et al. (1995) report 25% of players B choosing zero, while in this experiment 47% of players sent back zero at least in one decision in two rounds and only 4% (7 out of 178) players sent back zero for all received amounts in both rounds. The main dependent variable in all further results is the money sent back by player B as a fraction of money sent by player A, which takes values from 0 when player B keeps all the value, trough 1 when B sends back exactly the same amount as A chosen to send, to the maximum of 3, when player A returns all the money he received. Due to the use of the strategy method, each player B in every round makes 10 decisions for each possible amount received. Since these decisions taken individually are not independent, in the following statistical analysis the average of the above-mentioned 10 return rates is taken as one decision. In case of players A, the trust measure is the amount sent to player B as a share of the total money available, taking values from 0 to 1.
The summary of the general results is shown in Table II. The results show high levels of both trust and reciprocity among all groups of subjects. The obtained reciprocity levels are slightly
higher than averages from 168 investment games included in the meta-analysis made by Johnson & Mislin (2011). The share of the initial endowment sent by players A in the experiment was on average .64 – the mean in the meta-analysis was .5, with the highest value of .89 and the lowest of .22. Also, the rate of return on the amount sent by players A in this experiment is slightly higher in all samples than 1.11 – the average of the 137 experiments in the meta-analysis, from which .33 was the lowest reported value and 2.67 was the highest.
TABLE II
Mean share of available amounts sent to the counterpart by both types of players
Sample High School University Total
Second round type BONUS BONUS DICT -
Players A Average share .65 .60 .64 .64
(S.E.) (.02) (.03) (.05) (.02)
N 104 50 30 184
Players B Average share 1.23 1.15 1.58 1.27
(S.E.) (.05) (.07) (.11) (.04)
N 196 102 58 356
Note: The means shown in the table are the averages of 10 player’s B responses in each round. N reported are the numbers of observations – two observations per each player.
Basing on averages and medians of the amounts sent by players A, and the average responses to this amount by players B, the usual earnings were as follows: 11zł for player A and 23zł for player B in high school BONUS treatment, 10zł for player A and 22zł for player B at the university in BONUS treatment and 12zł for player A and 22zł for player B in treatment DICT.
4.2.1. Normality and use of statistical methods
Since the experiment’s focus is the reciprocal behaviour of players B, the following outcome analysis consider their role in the experiment, with only a few remarks made about players A. The distribution of the dependent variable, the share returned by players B, in all samples fails the test Shapiro-Wilk test for normality (p-value < .001). However, in case of a sufficiently big sample (around 100 observations), the parametric methods, such as t-test, do not require any normal distribution assumption to provide robust results (Lumley, Diehr, Emerson, & Chen, 2002). In the further empirical analysis, a parametric t-test, as well as non-parametric tests are
used. As the non-parametric method, Kolmogorov-Smirnov (K-S) test and Mann-Whitney U-test (MWU, also referred to as Wilcoxon rank-sum U-test) are used in data comparisons in case of independent samples – for example 1-1 and 1-4 sessions. In case of the dependent samples, which occur in control and BONUS/DICT rounds comparison, the Wilcoxon signed-rank (WSR) test for paired data is used. The reported statistics apply to two-sided tests unless specified differently.
4.2.2. Distribution among samples
A comparison of the distributions of the dependent variable across all samples is presented in Figure III. While BONUS treatments were organised with exactly the same procedures in high school and university, the samples slightly differed in observable statistics. In particular, the male/female ratio was higher in high school – 38% females in high school sample, 49% in university BONUS sample (MWU test rejects the same gender distributions with p-value < .001) and high school students were 3 to 4 years younger (18-19 years old in high school depending on class, 22 years old at university with only several outliers).
However, the distributions of decisions from both samples do not differ significantly. The Kolmogorov-Smirnov test fails to reject the hypothesis of the same distribution among these two samples with a p-value of .50, so does the Mann-Whitney U-test but with a p-value of .30. Looking at differences in every round (every combination of control/bonus and 1-1/1-4 treatments), neither of the abovementioned tests rejects the hypothesis of the same distributions between university and high school subjects. Since neither group statistically differs in distributions of the dependent variable, the high school and university BONUS samples are analysed together in further results. The potential effect of the aforementioned observable statistics is tested in regressions in the last part of the Results chapter.
Looking at the distributions in Figure III, the DICT treatment stands out from both BONUS treatments. Due to the different rounds designs, the choices of players B from DICT treatment can only be compared with BONUS treatments by looking for the differences in the control rounds. The hypothesis of DICT control rounds having the same distribution as BONUS control rounds cannot be rejected by both the MWU test value of .28) and by K-S test (p-value of .13). This suggests that the subjects, in general, behaved similarly in the standard investment game across all samples.
FIGURE III
Distributions of rate returned by players B in different samples
4.3.Social distance
To test whether a larger social distance decreases players’ reciprocity, the choices of players B from 1-1 and 1-4 rounds are studied. By comparing the means of all treatments together, players B returned less in the 1-4 treatments. The average return rate in all 1-1 sessions was 1.39, while in 1-4 sessions it was 1.18. The difference is highly significant (t-statistic: 2.60) and hypotheses of the same distributions in both types of sessions are easily rejected by Mann Whitney U-test (p-value < .01) and Kolmogorov-Smirnov test (p-value < .05). The more detailed analysis is shown in Table III. Due to insufficient sample sizes for the t-test, the significance levels marked in the table are counted by Mann-Whitney U-test.
TABLE III
Mean share returned by players B in all treatments, with marked differences between 1-1 and 1-4 sessions
Baseline Treatment Total
1-1 1-4 1-1 1-4 1-1 1-4 BONUS treatment Mean 1.34 1.15 1.33 1.07 1.33 1.11 (S.E.) (.09) (.07) (.09) (.08) (.06) (.05) N 62 87 62 87 124 174 Mean diff. .19 .26* .22** DICT treatment Mean 1.66 1.37 1.79 1.64 1.73 1.50 (S.E.) (.28) (.20) (.20) (.97) (.81) (.96) N 10 19 10 19 20 38 Mean diff. .28 .16 .22
Significance for Mean differences reported for Mann-Whitney U-test (*p < .05, **p < .01, ***p < .001) N represents number of decisions (= number of players) in each of compared groups
In both treatments, a larger social distance resulted in the decreased reciprocity of individuals. The total differences, which summarise results from treatment and baseline rounds within sample, are substantial in both treatments but significant only in BONUS treatment sample (t-statistic: 2.72). The MWU test rejects the same distribution hypothesis in this case with p-value < .01. Even though the total difference between 1-1 and 1-4 sessions is the same in DICT treatment as in BONUS treatment, a much smaller sample and a higher standard error in DICT makes this difference non-significant.
In both DICT and BONUS sessions, the baseline rounds show no significant differences between 1-1 and 1-4 sessions, contrary to what was hypothetically predicted. Taking together results from baseline rounds in both treatment types, results show 0.19 lower return in 1-4 sessions than in 1-1 (t-stat: 1.75, MWU p-value < .09), marginally significant at 10% level7. In case of treatment rounds, the BONUS treatment yields significantly lower return rates in 1-4
7
sessions compared to 1-1 (MWU p-value = 0.02), in line with predictions. The quasi-dictator DICT treatment rounds results were not significantly influenced by a bigger social distance. Summarising the comparisons, the results provide mixed evidence to the expectations of the H1 hypothesis. In BONUS sample, the results of treatment rounds turned out to be much more sensitive to the indirect relation between players, compared to control rounds. This may suggest that when strategic consideration is strongly emphasised, the players are more likely to shirk when the social distance is larger. While in BONUS treatment the general comparison between 1-1 and 1-4 sessions supports the hypothesis, in DICT treatment the general result is highly non-significant and finds no support to the theoretical predictions. A much smaller sample of DICT sessions can be the reason for such an outcome. An interesting fact is that in the treatment rounds of the DICT games, the reciprocity values stand out from all other treatments. In this quasi-dictator game, players B were sending back by far the most, regardless of how close the relation between the players was.
Considering the players A role, the average amount sent to players B (6.37zł) and median (7zł) was exactly the same in 1-1 and 1-4 sessions of BONUS sample, which consisted of 92 players A. This shows that players A were not affected by the larger social distance treatment. In DICT sessions, the sample size of 15 players A is too small to draw any conclusions, and the difference in the average amounts sent in 1-1 and 1-4 sessions (6.61zł in 1-1, 6.17 in 1-4) is highly insignificant.
4.4.Strategic consideration
The analysis of the impact of strategic consideration on reciprocity is slightly more complex, due to a twofold approach to test for its influence on behaviour. BONUS treatment included additional motivation to give strategically in the second round, whereas DICT treatment took the strategic consideration away in round one and restored it in round two. These two types of treatments yielded slightly different results.
4.4.1. BONUS treatment
The BONUS design was aimed to strengthen the reason for sending more money to players B. Players A indeed positively responded to the additional motivation, sending significantly more in the second round (mean of 5.88zł in control rounds and 6.83zł in treatment rounds, t-statistic: 2.59).
In case of players B, most of the students (51%) reacted to the existence of an opportunity for player A to earn an additional bonus by decreasing the amounts sent in treatment rounds.
However, a substantial group of players (38%) sent more in the round with a bonus compared to the control round. Looking at their strategies, self-reported in questionnaires, the motivation of the mentioned players was to let player A reach the threshold and earn an extra bonus. The differences between the mean share returned in the first and the second rounds are shown in Table IV. In 1-1 sessions, players B returned on average nearly the same amount in the round with bonus as in the control round. Looking at the decisions of players B in 1-1 sessions, 44% of them decreased the amounts in bonus rounds while 42% of players B increased the amounts sent back. The additional incentive for players A in sending money significantly decreased players’ B reciprocity in 1-4 treatments (WSR test for paired data p-value of 0.02). In case of 1-4 treatments, 56% of players B sent back less in bonus round and 31% of players returned more. This fact can be interpreted as a higher sensitivity to the selfish intentions when the social distance is larger than in case of direct 1-1 relation. However, this statement is not supported by results of the regressions discussed further. Therefore, a possible explanation can be that players B were more concerned about the payoff of players A (the fact whether A gets the bonus in particular) in 1-1 rounds, where the social distance was smaller, than in 1-4 treatment.
Taking differences in 1-1 and 1-4 sessions collectively, the two-sided WSR cannot reject the hypothesis of the same amounts sent in baseline and bonus rounds at 5% significance level. The same hypothesis can only be rejected by the one-sided test in favour of higher amounts sent in baseline rounds (mean 1.23) than bonus rounds (mean 1.18). This fact provides weak support to the hypothesis that strategic consideration decreases the reciprocity of players B. Summarising the abovementioned findings, results from bonus treatments fail to unequivocally support or reject the H2 hypothesis.
TABLE IV
Differences in mean share sent by players B between baseline and treatment rounds, divided by 1-1 and 1-4 sessions
1-1 1-4 Total
BONUS treatment Mean diff. .01 .08* .05†
N 62 87 149
DICT treatment Mean diff. - .14 - .26* - .22*
N 10 19 29
Significancereported for Wilcoxon Signed-rank test for paired data (†p < .1, *p < .05, **p < .01, ***p < .001) N represents the number of decisions (=number of players) in each of the compared groups
4.4.2. DICT treatment
In order to provide a more robust answer to the question of the influence of a strategic consideration on reciprocity, the DICT treatment was implemented. To provide less ambiguous results, DICT excludes the strategic consideration in the first round to examine the impact of including it in the second round. This approach indeed produced more vivid results concerning the H2 hypothesis. The quasi-dictator round resulted in the highest reciprocity levels among all 1-1 and 1-4 subsamples (the exact values are shown in Table III, Table IV displays the differences in means). The negative signs of mean differences reflect higher amounts sent in the treatment than in the baseline rounds. Although in 1-1 sessions the difference between the treatment and control round is non-significant, the difference in case of 1-4 sessions is relatively big and significant (WSR p-value <0.05). This goes in line with the H2 hypothesis. The problem with 1-1 session results is that these are based on the smallest subsample in the whole experiment – the decisions are made by only 10 participants from one session, almost two times less than the last but one smallest subsample8. Therefore, the inability to reject the hypothesis of the same medians is based on a very low-power measurement.
A general view on DICT results supports the theoretical predictions. In 1-1 and 1-4 sessions combined, 62% (18 of 29) of players B decided to send less in the round with strategic consideration than in the quasi-dictator round. The rest of the players sent the same or slightly higher amounts in the second round, with only one participant reporting that his strategy was to give more in round two. This outcome resulted in the significant, negative difference in distributions between two rounds (WSR p-value < 0.05). This goes in line with the H2 hypothesis stating that a strategic consideration of giving decreases the reciprocity of the receivers.
4.5.Additional analysis – regressions
To provide more insight on the effects of all potentially important variables, not included in comparisons, a series of regressions are reported. Due to heteroskedastic residuals, the regressions use robust standard errors. Standard errors are clustered on the players’ level to adjust the results to non-independent observations.
The regressions results are shown in Table V. The specification (1) shows results from both high school and university samples. In specification (2) the questionnaire reciprocity measures
are added as control variables9. That excludes the high school sample since the questions measuring reciprocity were added only in university questionnaires. In both reported regressions, the dependent variable is the share of the amount received that was returned by player B, just like in the rest of the former empirical analysis. The base level is 1-1 control round. The binary variables of interest are bonus (value of one in case of a BONUS treatment round), dict (1 in the quasi-dictator round) and 1_4 (1 if 1-4 session). To check for differences of treatment rounds between 1-1 and 1-4 sessions, the interaction terms bonus_4 (1 if 1-4 BONUS round) and dict_4 (1 if 1-4 DICT round) are included.
The coefficients of all variables of interest (bolded font in Table V) match the theoretical predictions. The negative coefficients of 1_4 variable follow the H1 hypothesis of the negative impact of a bigger social distance on receiver’s reciprocal behaviour. The coefficient is significant at 5% level only in the second specification (t-statistic 2.34), while in regression (1) it is marginally significant at the 10% level (t-statistic 1.72).
Considering variables specific to strategic consideration treatments, the negative value of
bonus coefficient show a decrease of reciprocity in rounds with stronger strategic consideration,
but the effect is not significant (t-statistics: (1) 1.05, (2) 1.56). Small and highly insignificant values of bonus_4 coefficients suggest no interaction effect of 1-4 and BONUS treatments combined. This contradicts one of the suggestions of previous comparisons that the negative effect of BONUS treatment on reciprocity is stronger in larger social distance sessions. In case of DICT treatment, the positive dict coefficients go in line with H1 predictions, indicating substantially higher values sent back in quasi-dictator rounds than in control rounds. The effect is significant at the 5% level in the specification (1), with a t-statistic of 2.16, but non-significant in specification 2. However, the interaction term dict_4 becomes much bigger and more significant in the second specification (t-statistics: (1) .2, (2) 1.65). Such situation is caused by the fact that university students are slightly more sensitive to the social distance treatment compared to high school students. This is reflected in a higher average return in university 1-1 control rounds and lower in 1-4 control rounds compared to high school10. This causes the difference between 1-1 control and 1-1 DICT rounds smaller and thus it becomes non-significant after excluding high school results. At the same time, the difference between
9 The six questions taken from SOEP (Wagner et al., 2007) are measuring positive and
negative reciprocity. Based on answers, the player’s positive and negative reciprocity is calculated. The questions and reciprocity calculation method is included in the Attachment B.
10 These differences are statistically insignificant due to K-S and MWU tests, as discussed in
1-4 baseline and 1-4 DICT rounds increases and is reflected in larger dict_4 interaction term, marginally significant at 10% level.
TABLE 5
Regression results: the dependent variable in both specifications is the share of the available amount sent back by player B
(1) (2) bonus - .113 - .252 dict .380* .203 1_4 - .206† - .420* bonus_4 - .005 .060 dict_4 .050 .317† gendera .082 .340* universityb .040 pos_rep .117* neg_rep .038 N 318 158
Sample High school & University University
Significancereported for t-test ( †p < .1, *p < .05, **p < .01, *** p < .001). N represents number of choices (2 choices per participant). Robust standard errors are clustered at player’s level.
a1=female b1=university sample
The reported results concerning variables of interest reflect the findings of the previous statistical analysis. Using the regressions, however, the ability to control for other observables arises. Considering potentially important factors, the positive gender coefficient shows that women have higher reciprocity levels than men, who are the baseline, yet this relation is significant only in case of university students11. The subjects from high school do not differ significantly from university students in reciprocity decisions, which is reflected by the small
11 Additional regression of results of high school students alone show no significant
and non-significant coefficient of university variable (t-statistic .38) in regression with two samples.
The general positive reciprocity level, calculated by summarising points from three questions examining this feature, has a significant, positive influence on the return rate of players B (variable pos_rep, t-statistic 2.09). The negative reciprocity is measured by the neg_rep variable and is calculated analogous to pos_rep. Looking at its coefficient, surprisingly more negatively reciprocal players returned higher share than less negatively reciprocal ones, but this relation is marginal and non-significant (t-statistic .84).
Using a different model, where all the player’s decisions were examined separately (the robust standard errors were also clustered at players’ level) allowed to examine whether higher amounts received caused higher share returned by players B. In numerous previous studies this relation was found not significant (including Berg et al., 1995). Also in the present experiment, the relation between the amount sent by player A and the relative share returned by player B turned out to be non-significant in regression and very weakly correlated (Spearman’s rank correlation coefficient of -.01, the hypothesis of independence between variables not rejected with a p-value of .52).
5. Discussion
In this part, the potential limitations of the research are discussed. Then the conclusions drawn from the research are summarised together with some future research suggestions and market implications.
5.1.Limitations
The study contains several limitations worth discussing. The first one is the sample size. While the sample of the BONUS treatment, both in 1-1 and 1-4 versions is satisfactory, the sample of the DICT treatment is substantially smaller. Such a situation is caused by the specific time the experiments at the UMCS university were organised – the ending of the academic year and the beginning of the exam period. The last day of regular classes, during which some of the DICT sessions were planned, was cancelled by the chancellor as a day off. Therefore, half of the potential DICT sample was lost.
The use of students as the subjects brings another possible issue. Several studies show that students participating in behavioural experiments show lower levels of reciprocity than non-student subjects (Falk, Meier, & Zehnder, 2010; Fehr & List, 2004). Nonetheless, considering the high overall levels of reciprocity in the studied sample, results do not seem to be biased towards selfish behaviour. Moreover, the differences in behaviour are studied in this research,
so the possible influence of this sample selection on reciprocity would affect only absolute levels in every treatment, not biasing the comparisons.
Another potential problem is an ordering effect, which arises when results in the second round are influenced by the first round. That could possibly occur especially in DICT sessions, where very high reciprocity levels in the quasi-dictator round could be the reason for relatively high reciprocity in the second, baseline rounds. One of the possible pieces of evidence of the ordering effect would be a significant difference in distributions between DICT baseline rounds and BONUS baseline rounds, which in turn were played as first in sessions. However, neither MWU nor K-S test rejects the hypothesis of the same distribution between baseline rounds. On the other hand, the possibility of an ordering effect influencing the results of BONUS treatment rounds cannot be excluded. A possible remedy for this problem could be reordering the rounds in some sessions. Since the specific number of sessions could not be fixed before the start of the experiments, reordering was not included for the fear of a too low number of each treatment and thus low power of measures. In case of comparisons between 1-1 and 1-4 treatments the independent samples were used, therefore the ordering effect cannot occur.
Several other design features are also worth commentary. All of the further discussed aspects apply to all participants equally, therefore they should not drive the comparison results. Due to financial constraints, in this experiment the random payment was implemented. Such a design means that (lack of) kindness or reciprocity in the decisions can have no monetary consequences when the player is not selected for the payoff, which in turn can decrease the perceived responsibility for the decisions. Nonetheless, the use of a random payment in the investment game was found in Johnson & Mislin (2011) meta-analysis to decrease the trust, but not to influence the reciprocity of players.
The use of the strategy method, according to some research, could cause lower perceived responsibility for the one’s actions towards counterpart (Güth, Huck, & Müller, 2001). Brandts & Charness (2011), in their literature survey, examined the effects of the strategy method on behaviour compared to the direct response method. The authors found that in most comparisons of the two methods, there were no significant differences in behaviour of the participants, and more importantly, in no case was the treatment effect observed by using strategy method unobserved with the direct response method. Also, the meta-analysis of Johnson & Mislin (2011) found no effect of the strategy method on the investment game results.
5.2.Conclusions and further research
Based on the results presented in the previous parts, several conclusions about the nature of reciprocity can be drawn. In general, the results provide strong evidence for the existence of reciprocal behaviour. Only 8 out of 178 receivers consequently sent back zero, as predicted by Nash equilibrium. Similarly, none of the senders saved all the amount in either round, which would suggest anticipation of selfish behaviour. These results reflect the findings of a wide range of previous studies proving the reciprocal nature of people, including (Berg et al., 1995) original investment game and numerous following experiments referred to in the Literature review.
The experiment findings undermine the outcome-oriented theory of fairness concerns. According to the outcome-oriented approach, the behaviour of receivers could be different only between baseline and BONUS rounds, since the additional sum of money can change the monetary outcomes. In contrast, the results show that BONUS has the smallest and least significant influence on players decisions among all treatments, whereas other treatments indeed impact players’ reciprocity. The fact that 1-4 and DICT treatments influence the behaviour of players cannot be explained using the outcome-oriented models alone. In this sense, the results support the earlier experimental findings of McCabe et al. (2003).
Another finding of the experiment is that relaxation of a direct connection between sender and receiver can substantially decrease the willingness of the latter to reciprocate, which was stated as the first research hypothesis. Receivers sent back lower amounts in 1-4 treatments compared to baseline 1-1 treatments, and in most comparisons, the difference is significant at 5% level. This may evidence that subjects feel less obliged to reciprocate kindness when the decision of sending some amount is not made individually towards them. Another, but not exclusive explanation, is that reciprocity is decreased by the knowledge that if one behaves selfishly, others can reciprocate to the same sender. Both these explanations indicate that people are more likely to shirk when the social distance is enlarged. In this way, the experiment finds support to the findings of Dana et al. (2007) and contradicts the results of van der Weele et al. (2014). The further research could attempt to separate the effects of the knowledge that the decision of giving was made towards a group of people from the possibility that multiple receivers can reciprocate the favour to the same sender, in order to derive pure influence of both features on reciprocity.
The results of the experiment also reinforce the approach that the intentions read by subjects play a great role in reciprocal decisions. The second research hypothesis, stating that strategic consideration decreases reciprocity of kindness receivers, obtained mixed evidence in
