Personal relations and their effect on behavior in an
organizational setting: An experimental study
how personal relations affect performance in organizations. In the game we use
a manager has to assign different degrees of decision power to two employees, who then
have to make decisions which affect the manager. Our evidence shows that managers favor
employees that they personally know and that these employees favor the manager in their
decisions. However, this behavior does not affect the performance of the employees that
do not know the manager. These effects are independent of whether the employees that
know the manager are more or less productive than those who do not know the manager
We present data from experiments in which we study whether people favor their friends in a stylized organizational
setting and whether this has any effect on the behavior of other people involved in the situation. Personal relations are
present in most organizations and are naturally prominent in family ﬁrms which recently have received much attention by
economists (see La Porta et al., 1999; Demsetz and Villalonga, 2001; Villalonga and Amit, 2006).
The issue of how the agency problem is modiﬁed in family ﬁrms is studied in Schulze et al. (2001),Burkart et al. (2003)
and Mork et al. (2005). A common theme of these studies is the owner’s altruism towards the heir and the possible effects
that this attitude may have in alleviating or worsening the agency problems. This has been analyzed theoretically by Chami
(2001)andBurkart et al. (2003). Favoritism towards relatives is one possible consequence of such altruism.
Similar problems of favoritism may arise in relation to personal friends in other kinds of organizations. Westphal and
Stern (2006)show that personal relations play an important role for managers in order to get better positions at their ﬁrms,
substituting credentials. In a study on ﬁrms listed on Forbes 500, they show that ingratiatory behavior directed at the CEO
pays off, increasing the likelihood to receive better appointments. Kristof-Brown (2000)reports that social relations appear
as the primary force driving appointments into management positions. Hence, discriminatory practices based on personal
relations may go beyond family ﬁrms and reach other organizations as well.
The consequences of such favoritism are somewhat controversial. Fama and Jensen (1983)consider that family relation-
ships between owners and managers reduce the agency problem.
Miller and Le Breton-Miller (2006)discuss the possible
positive effects of choosing a family member as a top executive due to higher motivation—in the line of stewardship theory.
Other analysts provide arguments and evidence in favor of the effects of favoritism being negative. Pérez-González (2006)
analyzes CEO successions in family ﬁrms using data from U.S. corporations. He ﬁnds that promotion of CEOs with family ties
is indeed frequent. Family CEOs attain this position earlier than non-family CEOs—on average eight years earlier. This kind of
pro-family bias leads to considerable declines in performance, as measured by returns on assets and market-to-book ratios.
Kets de Vries (1993)andSchulze et al. (2001)indicate that favoritism may also affect other decisions besides CEO succession:
promotion to selected places in the organization, better remunerations or more training. Kole (1997)andBates et al. (1998)
ﬁnd evidence of these practices. We will refer to the bias based on family or personal relations and to the possible reactions
– positive or negative – of favored individuals as the direct effects of favoritism.
However, this may not be all that needs to be considered. Favoritism may also have an indirect effect. If a decision
based on family relations or friendship violates economic or fairness principles, other agents in the organization may feel
discriminated since they do not beneﬁt from the decision only because they do not belong to the insider group. Hence, they
may perceive that their earnings, positions, status or job security levels are below what they deserve in comparison to the
insiders. If this perception arises, negative reactions from these agents may result, as Gomez-Mejia et al. (2001)andMiller
and Le Breton-Miller (2006)point out.
In this paper we use a simple experimental design – based on the well-known trust game – to study if, in the presence
of personal relations between agents, favoritism occurs and whether it leads to poor results.2The essence of what goes on
in the organizational settings we are interested in can be captured in a situation in which a decision-maker or manager has
to assign certain unequal degrees of responsibility or decision power to two different subordinates or employees; one of
the employees will obtain discretionary power over a larger part of a pie than the other. These two employees then have to
make distributive decisions which affect themselves and the manager.
We study the impact of two treatment variables. The ﬁrst is the existence of a personal connection between the manager
and one of the employees; we compare the cases where this connection exists and where it does not exist. The other
treatment variable is the ratio of efﬁciency factors of the two employees. In the base case the two employees have the same
efﬁciency factors, whereas in the two other cases we look at the efﬁciency factors are different. The resulting design with
six treatments makes it possible to study the effect of a personal relation between the manager and one of the employees
for the cases where the employee in question is equally, more or less productive than the other employee.3
Our treatment conﬁguration makes it possible to study both the direct and indirect effects of favoritism. In designing the
experiment we have aimed at ﬁnding an unbiased environment, which a priori does not seem to be especially favorable,
neither to the existence nor the absence of the two effects of favoritism. Our objective was to let the data speak for themselves.
We ﬁnd evidence for the direct favoritism effect in that managers tend to favor employees that they personally know
and that these employees tend, more than other employees, to favor the manager in their distributive decisions. However,
favoritism does not affect the performance of the employees that do not know the manager. All these effects are independent
of whether the employees that know the manager are more or less productive than the employees who do not know the
manager. We also ﬁnd that the joint performance of employees is signiﬁcantly higher in the presence of personal relations,
regardless of efﬁciency differences.
2. Design and hypotheses
In our game Player A is given a ﬁxed amount of 10D that has to be passed on to Player B and Player C in ﬁxed proportions.
The choice that A has is to give 6D to B and 4D to C or vice versa.4
Player A can be seen as representing the manager of a ﬁrm or organization who has to assign different levels of respon-
sibility or decision power – represented by the different shares of the initial pie – to two of her employees, B and C.
Once Player A has decided how to assign the two shares, the amount sent to Player B is multiplied by a factor MB and the
amount received by Player C is multiplied by MC. These factors represent the efﬁciency levels of the two agents or employees.
The introduction of these factors allows us to distinguish between responsibility and efﬁciency levels. While the shares the
employees are assigned by the manager can be seen as the responsibility levels, the multiplicative factors can be naturally
interpreted as the productivities of the two employees. This distinction is important for our analysis below. The game we
use is a variant of the trust game, in which Player A has to trust fully, but can decide whom to trust.
Our design incorporates two treatment variables. The ﬁrst of them pertains to whether the principal personally knows
one of the agents or not. In our anonymous treatments the game was played in the standard fashion: subjects did not know
who the others in the trio were. In our friends treatments Player A and Player B knew each other personally, but both did
not know Player C personally. In turn, Player C did not know either Player A or Player B, but did know that Players A and B
knew each other personally.
Apart from varying whether A and B know each other or not, we also vary the factors MB and MC. We study three different
pairs of efﬁciency factors. The ﬁrst is the case of equal efﬁciency factors, MB=MC= 3. The pies that – after the multiplication –
the two employees have to distribute will be of sizes 18D and 12D
In the second conﬁguration of efﬁciency factors, Player B
is a low efﬁciency employee (hereafter we will refer to this case as B Low) and the two efﬁciencies are MB = 2.5 and MC = 3.75.
In the third case B is a high efﬁciency agent (hereafter, B High): MB= 3.5 and MC= 2.25.6
The numbers for the B Low and B High cases were selected in a way that keeps the total pie size equal to the one in the
baseline case of equal efﬁciency factors at 30D for the case in which B obtains the larger share.7Given this restriction B and C
players cannot simply exchange the efﬁciency factors between the two unequal efﬁciency treatments. The chosen efﬁciency
factors satisfy the restriction given by the baseline; in addition we tried to keep the ratio of the efﬁciency factors as close
as possible to the ratio of shares. The ration of shares is 1.5, since (high share)/(low share) = 6/4, while the ratio of efﬁciency
factors is MB/MC= 3.5/2.25 = 1.55 for the B High case, and MC/MB= 3.75/2.25 = 1.5 for the B Low case. However, remember
that the comparisons that we are mainly interested in are the ones corresponding to the friends vs. anonymous distinction,
so that the above choice of parameters is not crucial.
Once Players B and C have been informed about the amount received they have to decide simultaneously and without
any communication how much of the pie they control they want to give (back) to Player A; we call these amounts xBand
xC. The payoff for player A is the sum of the amount sent by Players B and C. Players B and C get, respectively, what they
decide to keep for themselves. The straightforward game theoretical prediction in this game – if players’ utility functions
just incorporate their own payoff, the game is played only once and players do not know each other – is that both Player
B and Player C will give nothing to Player A. Player A is, hence, indifferent with respect to how to distribute the initial pie
between Player B and Player C.
Table 1presents a summary of our treatments. In total we have six different cases which arise from the three different
efﬁciency conﬁgurations of Player B and C and the existence or absence of personal relations between Player A and Player
B. Table 1also introduces the acronyms for the different treatments which we will use below.
Our simple set-up makes it easy to focus on the precise issues that we want to study. We can separately analyze the
behavior of Players A, B and C and we can study the impact of efﬁciency factors on behavior. First, we are able to study how
player A’s assignment decision is inﬂuenced by whether she personally knows Player B.
Second, we can study how Player B’s decision is inﬂuenced by whether he has been chosen by an A that knows him. In
studying Players A and B in the friends treatment one has to take into account precisely that they know each other, so that,
one could say, that the experimental interaction will continue after the experiment itself. In particular, the two friends may
be able to share payoffs once the experiment is over. We will get back to this when discussing the results.
Third, we can study the reaction of the C player to the existence of a personal connection between A and B. For all three
cases we will be able to see how the inﬂuence of personal relations is affected by the efﬁciency differences between B and
C. Note that the behavior of Players A and B pertain to the direct effects of favoritism, whereas it is in C’s behavior where we
may ﬁnd indirect effects of favoritism.
2.1. The direct effects of favoritism
The ﬁrst issue is whether A players will tend to give the larger share of the pie to Bs that are friends. This tendency is
what we refer to as favoritism. Altruism has been argued to be one of the main motives behind the direct effect of favoritism.
Most people care more about family members or friends than for others and try to help them in the organization.
Of course, favoritism may also be inﬂuenced by the strategic element of expecting to get more back from a friend than from a stranger.
Both forces can go together since it is possible that the altruism is mutual.
Note, however, that it is not obvious what to expect. A could suspect that a C who is favored vis-à-vis a B friend could
be especially generous. In addition, the decision could be affected by the difference in efﬁciency factors. When the friend is
the more productive of the two employees simple favoritism should bias the A player even more towards giving the larger
share to B. Of course, it is also conceivable that more general fairness motivations may guide A’s behavior in the opposite
direction to compensate the C player for his low efﬁciency factor.
It is not easily foreseeable how these factors will interact. However, minimal groups’ experiments (see Yamagishi et al.,
1998for a survey) show that agents send more money to in-group partners because they expect more reciprocity from the
group members. Several papers study ingroup outgroup effects in the context of trust games. Glaeser et al. (2000)ﬁnd that
social connectedness – as captured by the number of common friends and the duration of the acquaintanceship – generally
predicts the levels of trust and trustworthiness in the game.
Falk and Zehnder (2006)report evidence from a cross-city-
district ﬁeld experiment in Zurich which shows that people trust strangers from their own district signiﬁcantly more than
strangers from other districts. Burnham et al. (2000)report that trust and trustworthiness is affected by whether the person
that an individual is matched with is labeled as “partner” or as “opponent.” Hence, this literature suggests that discrimination
arises as a consequence of expected reciprocity. At this point we posit what we call the favoritism–discrimination hypothesis.
Hypothesis 1 (Favoritism–discrimination). Principals in the friends’ treatment will give the higher share to their friend,
independently of efﬁciency factors.
The second part of the direct effect of favoritism has to do with whether B will react to being favored in a way that is
favorable to A. In the experimental literature in economics there is considerable evidence of reciprocal behavior in situations
like the one we study (e.g. Berg et al., 1995, etc.), but not on whether this is affected by the existence of personal relations.
We propose that social distance will affect reciprocal behavior.
Hypothesis 2 (Social distance decreases reciprocal behavior). Friends that obtain the higher share will give back more (in
percentage) than anonymous players
Note that the data could be consistent with only one of the hypotheses. If Hypothesis 1were upheld by the data,
but Hypothesis 2not, this could be described as “thankless friends” not reciprocating the good treatment they receive.
If Hypothesis 2were consistent with the data, but not Hypothesis 1, then managers would be getting something for nothing
from their friends. Both these possibilities did not appear very likely a priori, given the ample evidence for reciprocal giving
in games like the one we study here.
2.2. The indirect effect of favoritism
The idea of workers reacting negatively to favoritism, suggested by Schulze et al. (2001)andGomez-Mejia et al. (2001)
can be supported by several theories. Intentionality arguments suggest that if an agent perceives that he is being treated
unfairly, this may prompt negative reactions (Rabin, 1993).8The perception of unfair treatment can have several origins.
Equity theory (Adams, 1965), for example, claims that agents value what they obtain with respect to what they contribute.
This means that agents that perceive that they deserve more than they obtain will become angry.
Social identiﬁcation can also provide a basis for this negative reaction to discrimination based on the membership to
social groups. When identiﬁcation is not possible because one person is in a disadvantageous social position, resentment
may emerge and cooperation will suffer (Tajfel and Turner, 1986).Milton and Westphal (2005)ﬁnd evidence of this behavior
in work groups. All these ideas led us to expect a negative reaction to discrimination, as captured in Hypothesis 3.
Hypothesis 3 (Discrimination leads to negative reactions). Anonymous agents, when they are less favored in the friends’
treatment will give back less than in the anonymous treatment when they are less favored.
It is not self-evident how the differences in efﬁciency factors will impact on behavior. However, the discussions in Gomez-
Mejia et al. (2001)andMiller and Le Breton-Miller (2006)suggest that the reaction of the anonymous player will be more
negative when the manager’s friend is less able than the other employee.
3. Experimental procedures
Our experimental sessions took place at the Universitat Autònoma de Barcelona between April and October 2005. The
total number of participants was 429. The sessions were hand-run and involved two large classrooms, which we will refer to
as the A and the B/C room. To recruit participants we posted announcements at different locations on campus, in which we
asked interested subjects to sign up in a particular ofﬁce. We posted separate announcements to recruit subjects for the two
rooms. The announcements corresponding to the two different rooms were posted in different buildings of the university
in an attempt to minimize prior contact between subjects in the two rooms. At the moment of signing-up participants were
told in which room to gather for the experiment.
There were small but important differences between the anonymous vs. friends sessions, both with respect to what the
announcements said and to how the sign-up process was conducted. For the anonymous treatments the announcements
for the two rooms were identical. For the friends treatments the announcements for the A room asked for participants that
wanted to take part in the experiment in pairs, while the announcements for the B/C room were just like those for the
anonymous treatments and did not make any reference to pairs. Pairs that participated in the experiment had to sign up
together at an ofﬁce on campus.
When signing-up in the ofﬁce for the anonymous treatment participants simply wrote their name on a sheet either for
the A room or for the B/C room. We asked a certain number of subjects to gather in the B/C room and asked double that
number of subjects to gather in another classroom, the A room.
In the friends treatments there was a difference between signing-up for the A or the B/C room. Signing-up for the B/C
room took place in exactly the same way as for the anonymous treatments. In contrast when signing-up for the A room
subjects had to do it in pairs of people who personally knew each other. More speciﬁcally, pairs had to write their names at
the same time on the inscription sheet for the A room.
For both treatments, the subjects who gathered in the B/C classroom were automatically assigned the role of C. For the
subjects that initially gathered in the A classroom we used a random procedure to determine who would have the A and who
the B role. For the anonymous treatment subjects simply drew lottery ticket which assigned half of them to the A role and the
other half to the B role. For the friends’ treatment the assignment of the A and B roles took place as follows. For each pair of
subjects that had signed up together for the session we separately determined randomly who would be A and who would be B.
Subjects that had been assigned the Player A role stayed in the A room, whereas the B players where guided to the B/C
classroom. While in the A room the A and B roles were being assigned, in the B/C classroom subjects with the role of Player
C had been randomly seated in order on the right side of the room. Once the B players arrived there they were seated on the
left side of the classroom. In the anonymous treatment they were randomly assigned to seats. In the friends treatment the
B players were seated in an order that facilitated keeping track of the pairings with the A players.
Then instructions were read aloud, simultaneously in both rooms. After we had ﬁnished reading the instructions and
answered questions, we distributed decision sheets to subjects.9Again, the friends treatments involved a particular feature:
the decision sheet of each A (B) player showed the name of the B (A) player they were paired with. In the anonymous
treatments the decision sheets simply showed the identiﬁcation number of the other two people in the trio. This identiﬁcation
numbers could not be tracked by participants to any other individual in the session.
The A players moved ﬁrst; they marked their decision – whether to give 6D to B and 4D to C or vice versa – on their
decision sheet. We then marked the decision on the corresponding decision sheets of the B and C players in the B/C room.
Then the B and C players made their decisions and these were communicated to the corresponding A players. That was the
end of the experimental session; the decisions were one-shot in character.
Table 2presents the proportions in which players B in our experiment obtain the higher share in the different treatments.10
The pattern is remarkably simple: for all three anonymity cells there is a small bias towards B, perhaps caused by the fact
that A and B were initially together in the A room or by the fact that B precedes C in the alphabet. In contrast, in all three
friends cells there is a similar large bias towards B.
This impression is conﬁrmed by a Chi-square test which compares the differences in proportions between the three cases
involving friends and the three involving anonymous B players; the p-values for the three pair-wise Fischer exact tests are
.038, .004 and .003. In short, with respect to Hypothesis 1we ﬁnd that favoritism exists on the side of Player A and is the
same irrespective of the efﬁciency factor of the friend.11
Table 3presents the mean returns that Player A obtains from Player C and Player B in those cases in which Player C (who
in the treatments with friends is the anonymous player) gets the low share and Player B gets the high share. With this data
we can discuss Hypothesis 2(Social distance decreases reciprocal behavior) and Hypothesis 3(Discrimination leads to negative
Hypothesis 2states that friends will give back more (in percentage) than anonymous players. We observe that the returns
from Player B are systematically higher in the friends’ treatments than in the anonymous treatments. Like the behavior of
the A players, the pattern is quite independent of efﬁciency differences. In treatment FEQP this average return is 43 percent,
whereas in AEQP it is 23.8 percent. Using the Wilcoxon rank-sum test, this difference is strongly signiﬁcant (p = 0.0002).
This result appears also in treatments FBLOW and ABLOW. In FBLOW the average return is 41.01 percent and in ABLOW
the return from Player B is 22.22 percent, an again strongly signiﬁcant difference (p = 0.0023). Finally, in FBHIGH the return
from Player B is 36.72 percent, while in ABHIGH is 19.19 percent, a signiﬁcant different at p = 0.024. Hence, we conclude that
Hypothesis 2is conﬁrmed.
Players A and B that knew each other could conceivably share their money payoffs after the experiment; in a way this
just reﬂects how things are in the situations that we are trying to represent. However, we think that payoff-sharing cannot
explain away the fact that B players that are friends give back more than strangers. In other words we do not believe that
after the experiment the B players in the friends treatments asked the corresponding A players to pay them back what these
As had obtained during the experiment in excess of what the As in the anonymous treatments had obtained. Perhaps the
B players were asked by their A friends to pay them the rest up to the 50 percent sharing, but our impression is that this
did not happen. The B players got away with keeping around 60 percent of the amount to be shared. In general, subjects
accepted the framing of the situation into which we had put them.
We now turn to the returns from players of type C when they get the low share. Hypothesis 3stated that players of
type C would return a lower amount in those treatments where she obtains the low share in a group of friends than in the
anonymous treatment. Our results show that this is not so and, again, this fact is not affected by efﬁciency differences. In
FEQP the return is 13 percent, whereas in AEQP it is 15.87 percent, consistently with this idea, but this difference in not
signiﬁcant at any conventional signiﬁcance level. In treatment FBLOW the mean return from Player C is 16.37 percent, while
in ABLOW is 11.11 percent; the difference is not signiﬁcantly different. Finally, in FBHIGH the mean return from Player C is
8.76 percent and 5.24 percent in ABHIGH. Once more, this difference is not signiﬁcant. Hence, Hypothesis 3is clearly rejected
in our design. Surprisingly, players of type C do not appear to be bothered by the fact that the other player, a friend of Player
A, gets the high share, even in situations where the friend is a low efﬁciency agent.12
We can now make some additional remarks about the results. Notice ﬁrst that, as an implication of the above results, the
total return – shown in the last column of Table 3– is signiﬁcantly higher for treatments where friends play, and that this
occurs regardless of the efﬁciency factors. In FEQP the total return to Player A is 30 percent vs. 20.63 percent in treatment
AEQP (Wilcoxon rank-sum, p = 0.011). In treatment FBLOW the total return amounts to 28.01 percent of the resources,
almost twice as in ABLOW, 14.58 percent (Wilcoxon rank-sum, p = 0.0007). Finally, total return in treatment FBHIGH is 28.53
percent vs. 13.13 percent in ABHIGH (Wilcoxon rank-sum, p = 0.0007). We will get back to this result at the end of the next
5. Summary and conclusions
Our ﬁrst result is that, given the opportunity to choose between a friend and an anonymous player in a trust relationship,
subjects in the manager role discriminate – in a statistical sense – in favor of the friend. This discrimination occurs inde-
pendently of whether the friend has a higher or lower efﬁciency factor than the anonymous player. Second, we ﬁnd that
friends return systematically more than players in parallel anonymous treatments, again with no effects due to differences
in efﬁciency factors. These results are consistent with the view that personal relations help mitigating moral hazard prob-
lems in environments characterized by contractual incompleteness, an idea discussed informally in the literature on agency
problems as in Fama and Jensen (1983)orMiller and Le Breton-Miller (2006). In a more general sense, our results are in line
with the notion that social preferences act as contract enforcement devices, as in Fehr et al. (1997).
Our third result is that anonymous players that interact in a game with two friends and obtain the smaller share of the
pie (and in this sense are discriminated) do not react by lowering their return to the principal with respect to analogous
situations in a purely anonymous game. This is again independent of whether the friend is more or less efﬁcient than the
third player. This surprising result can be understood in terms of, what in our context, are the relevant social comparisons
(see Akerlof, 1982, 1997).Tropp and Wright (1999)argue that an individual’s sense of relative deprivation may depend on
whether the comparison is perceived to be inter-group or intra-group, and on the level of identiﬁcation of the person with
respect to his own group. They ﬁnd that in self-outgroup comparisons, highly identiﬁed individuals report more deprivation
than individuals low in identiﬁcation. In this sense, Miller (2001)points out that people believe that they deserve more
respect from other individuals pertaining to the same group.
The lack of the third player’s reaction to the statistical discrimination we observe can be understood in terms of a self-
outgroup comparison; the third player may naturally consider himself as part of a different group than the two friends.
Another relevant issue here may be legitimacy. Tyler (2006)argues that when differences between groups are legitimate,
people will not perceive bad outcomes as a reﬂection of discrimination. On the contrary, if differences between groups are
perceived as illegitimate, they may generate anger. This effect has been observed by psychologists in experimental research
(Brown and Ross, 1982, among others). Applied to the case of family ﬁrms it suggests that non-family members may interpret
that family members have a legitimate status in the family ﬁrm and, hence, they accept discrimination.
Our fourth result brings the previous results together and can be seen as a kind of bottom line of the whole study. The
manager’s total earnings are higher in the case where he interacts with B being a friend than when B is an anonymous
player. From the manager’s point of view, it is better to be in an experimental ﬁrm with friends, and it is perfectly rational
to trust the friend more than the stranger. In addition, the earnings of the manager are very similar across the three cases
with friends (FEQP, FBLOW and FBHIGH). To see how this comes about just compare behavior between FBLOW and FBHIGH
in Table 3. The discriminated C player in FBHIGH gives back less in absolute terms and in proportions than in FBLOW. At the
same time the friend makes a somewhat larger transfer when he is less efﬁcient – in FBLOW – than in the other case, leading
to the overall result. This result suggests that personal relations could indeed be efﬁciency-enhancing in situations where
agents’ decisions are not purely distributive, but also productive.
Financial support from Spanish Ministerio de Ciencia e Innovación (SEJ2007-67895-C04-03 “Organización de la empresa,
prácticas de gobierno y control familiar”), the Barcelona GSE Research Network Recognition of Research Program, Consolider
Ingenio 2010 (CSD2006-00016) and the Banca March Familiy Business Chair is greatfully acknowledged. The paper beneﬁtted
from presentations at the ESA Asian regional meeting in Hong Kong and the ESA international meeting in Atlanta, both in
2006. The authors thank David Rodríguez and Javier Valbuena for help in running the experiments.
مطالب مشابه ...