doing good or doing well? image motivation

8/14/2019 Doing Good or Doing Well? Image Motivation

1/32

Doing Good or Doing Well? Image Motivation and

Monetary Incentives in Behaving ProsociallyDan Ariely, Anat Bracha , and Stephan Meier

January 22, 2008

Abstract

This paper experimentally examines image motivationthe desire to be liked andwell-regarded by othersas a driver in prosocial behavior (doing good), and askswhether extrinsic monetary incentives (doing well) have a detrimental effect on proso-cial behavior due to crowding out of image motivation.

By denition, image depends on ones behavior being visible to other people. Usingthis unique property we show that image is indeed an important part of the motivationto behave prosocially. Moreover, we show that extrinsic incentives interact with imagemotivation and are therefore less effective in public than in private. Together, theseresults imply that image motivation is crowded out by monetary incentives; which inturn means that monetary incentives are more likely to be counterproductive for publicprosocial activities than for private ones.

Keywords : prosocial behavior, extrinsic incentives, image motivation, experiments

JEL : D64, C90, H41

Sloan School of Management, Massachusetts Institute of Technology.The Eitan Berglas School of Economics, Tel Aviv University.The Federal Reserve Bank of Boston.

1


2/32

1 Introduction

Most charitable organizations depend on private contributions, in the form of monetary gifts,

volunteer efforts, or other tangible contributions, such as blood donations. The magnitude

of private contributions is impressivein the United States alone 89 percent of households

donate, averaging $1,620 per year, and 44 percent of US adults volunteer the equivalent of 9

million full-time jobs (Independent Sector, 2001). This prosocial behavior is striking in light

of the economic incentive to free-ride in the provision of public goods. In order to elicit con-

tributions, charitable organizations use many creative efforts to incentivize voluntary giving.

For example, the Lance Armstrong Foundation recently launched the LIVESTRONG cam-

paign where individuals contribute by purchasing colorful wrist bands, many organizations

give thank-you gifts, such as mugs and calendars, and charities organize walks, concerts, and

advertise lists of their donors. In addition, the government helps increase charitable giving

by offering tax breaks for such donations.

The various types of charitable contributions and the many real-life ways of soliciting

such donations suggest that there may be different motives for individuals to behave proso-

cially. These motives are roughly divisible into three broad categories: intrinsic, extrinsic,

and image motivation. Intrinsic motivation is the value of giving per se, represented by

private preferences for others well-being, such as pure altruism or other forms of prosocial

preferences (for surveys, see Fehr and Schmidt, 2003; Meier, 2007). Extrinsic motivation is

any material reward or benet, either monetary or non-monetary, associated with giving,

such as thank-you gestures and tax breaks. Image motivation , or signaling motivation, refers

to individuals tendency to be motivated partly by how others perceive them. Image motiva-

tion therefore captures the rule of opinion in utility, i.e., the desire to be liked and respected

by others and by the self. If individuals are looking for social approval in their behavior,

then they should try to signal traits which are dened as good based on social norms and

values (for economic models incorporating social approval, see e.g., Akerlof, 1980; Hollander,

1990; Benabou and Tirole, 2006; Andreoni and Bernheim, 2007; Ellingsen and Johannesson,

2


3/32

2007, 2008). Being altruistic is often seen as good, and being greedy or selsh is not.

Prosocial behavior is therefore a way to signal to others that one is good. 1

The desire for social approval implies that, conditional on prosocial activity yielding a

positive image, people will act more generously and prosocially in public than in privatesettings. A number of eld and laboratory studies have found such a pattern (e.g., Andreoni

and Petrie, 2004; Dana et al., 2006; Rege and Telle, 2004; Soetevent, 2005). This may explain

why many organizations make individuals contributions explicitly visible to others by having

charity walks, offering wrist bands, posting lists of donors, etc. Indeed, very few donations

are actually done anonymously (see Glazer and Konrad, 1996).

While these three motivations (intrinsic, extrinsic, and image) have separate effects on

prosocial behavior, they may also interact with each other. Following Titmuss (1970), who

claimed that private monetary incentives for blood donations would decrease blood supply,

a number of empirical studies have shown that private monetary incentives indeed have

negative effects on prosocial behavior (e.g., Frey and Oberholzer-Gee, 1997; Gneezy and

Rustichini, 2000a,b; Mellstroem and Johannesson, 2008). While there are a few studies

showing detrimental effects of extrinsic incentives, less is known about the mechanisms by

which this occurs. A number of explanations have been offered: extrinsic motivations might

interact with the intrinsic motivations in the decision to undertake a task (e.g., Deci, 1975;

Frey, 1997); extrinsic incentives might destroy trust in a principal-agent relationship, leading

the agent to behave less prosocially (e.g., Falk and Kosfeld, 2006; Fehr and List, 2004; Fehr

and Falk, 2002); or the introduction of extrinsic motives might shift individuals decision

frame from a social frame, in which prosocial behavior is more prevalent, to a monetary frame

in which maximization of self-interest becomes more salient (e.g., Gneezy and Rustichini,2000a; Heyman and Ariely, 2004).

This paper experimentally tests another mechanism by which extrinsic incentives can1Behaving prosocially can also be undertaken instrumentally in order to signal ones wealth or status or

in order to get future extrinsic rewards, like a political career or admission to a college (e.g., Glazer andKonrad, 1996; Harbaugh, 1998a,b; Blumkin and Sadka, 2007).

3


4/32

have detrimental effects on prosocial behavior: extrinsic incentives might interact with image

motivation by diluting the signaling value of prosocial behavior (see, Benabou and Tirole,

2006). Simply put, when observing a prosocial activity, the addition of extrinsic incentives

makes it more difficult to answer the question: Is the individual behaving prosocially to dogood or to do well? This will affect the image gain from prosocial behavior. Think of the

following scenario: an individual is considering buying a new hybrid car. This car is more

expensive than an equivalent car operating with a standard gasoline engine, but the hybrid

car helps in preserving the environment. Driving around a car which is clearly a hybrid car

would probably add to ones positive image, especially if she lives in a community that values

environmentally-friendly technologies. Suppose now that the government gives a large tax

benet for those who decide to purchase a hybrid car (and everybody knows about this).

The tax incentive, of course, reduces the price of hybrid cars and therefore should make

the hybrid car more attractive for the individual. However, the tax incentive also decreases

the image value of driving the hybrid car. Without the tax incentive, buying the hybrid

car denitely shows the individual cares for the environment (positive image), with the tax

incentives, it does not. More generally, when extrinsic incentives are provided (such as a

tax benet), it is difficult to conclude whether the prosocial act is due to ones innate good

traits (the persons concern for the environment) or due to greed (receiving a tax benet).

Therefore, if image indeed motivates prosocial behavior, introducing extrinsic rewards may

reduce image motivation, which can lead to a negative net effect (relative price effect net of

crowding out effect) on prosocial behavior.

To test for image motivation and its interaction with monetary incentives we develop

two behavioral hypotheses. The hypotheses make use of the different properties of thedifferent incentives driving prosocial behavior. Namely, intrinsic and extrinsic motivations

are personalaltruism, as an example of intrinsic motivation, is a personal taste, while

monetary payment and gifts, as examples of extrinsic motivation, are personal material

rewardsyet image motivation is a social motivation that depends on the visibility of ones

4


5/32

actions. Importantly, the interaction between extrinsic incentives and image motivation

depends on the prevalence of image motivation and therefore on the visibility of the prosocial

behavior. In anonymous prosocial decisions, for instance, extrinsic monetary incentives will

not crowd-out image motivation, as in this setting there is no image motivation to beginwith. 2 If extrinsic motivation does crowd-out image motivation, we expect that extrinsic

incentives will be more effective in private decisions to behave prosocially than in decisions

which are publically visible (Effectiveness Hypothesis). Consequently, extrinsic incentives

are more likely to be counter-productive in public rather than private decision-making.

We conduct an experimental study to investigate whether image considerations indeed

play a part in motivating prosocial behavior, and how this motivation interacts with extrin-

sic monetary rewards. Our test of the interaction between extrinsic and image motivation

uses the unique property of image concerns its dependence on visibility. More specically,

in the main experiment, named Click for Charity, subjects could donate to a charitable

organization by clicking two keys on the keyboard. Participants were randomly assigned to

make their donation choices in a public or in a private setting, and were randomly assigned

to receive, in addition to the donation made on their behalf, monetary incentives that ac-

crued solely to them. In addition, they were randomly assigned to one of two charities (the

American Red Cross or the National Rie Association (NRA)) which were either associated

with positive or negative image consequences. In the main analysis we focus on the good

charity as it is a clear prosocial activity. The good charity is dened either according to the

consensus at Princeton (the Red Cross), or according to participants perception of the ma-

joritys view on campus. The results strongly support the hypotheses that image motivation

is important for prosocial behavior, and that private monetary incentives partially crowd-outimage motivation. That is, without monetary incentives we nd signicantly greater effort

in public than in private; when monetary incentives are introduced, monetary incentives

have no effect on effort in public, while they do increase effort in private, and this differen-2There may be self-image which exists also in anonymous giving. However, here we take image to be the

value added by others opinion, and not of one own opinion.

5


6/32

tial effect is statistically signicant. The results for a bad cause with a negative image

valuedened either according to the consensus at Princeton (the NRA), or according to

participants perception of the majoritys viewsupport the partial crowding out hypothesis

as well. That is, introducing monetary incentives have no effect on contribution effort inpublic, while monetary incentives do increase the contribution effort when the decision is

private. We were able to further examine some of these hypotheses in a eld study (which is

more limited due to our setup), and in general the results support what we nd in the lab:

private monetary incentives seem to interact negatively with image concern, leading to the

result that monetary incentives are more effective in motivating private prosocial decisions

than ones made in public.

The paper proceeds as follows: Section 2 presents the behavioral hypotheses. Section 3

presents the experimental design, and section 4 the results. Section 5 presents the illustrative

eld study. Section 6 discusses the implications of our ndings and concludes.

2 Behavioral Hypotheses

This section develops behavioral hypotheses for our experimental setting to test for imageas a driver in prosocial behavior and to test whether extrinsic monetary incentives crowd-

out image motivation. An individuals incentives to behave prosocially can be divided into

three basic motivations: intrinsic, extrinsic, and image motivation. Intrinsic motivation is a

personal taste for giving per se, extrinsic motivation is any personal material reward, either

monetary or not. Image motivation is inuenced by what other people think. There are

many personal traits that yield positive image: being prosocial, fair, or caring; while being

unfair, or greedy, reduces or even yields negative image (e.g. Benabou and Tirole, 2006;

Andreoni and Bernheim, 2007; Ellingsen and Johannesson, 2008).

We focus on the case where individuals image is a function of how prosocial or greedy

other people think she is, as in Benabou and Tirole (2006). 3 The more people think one3In the Appendix we formally present the model of Benabou and Tirole (2006), and show how to derive

6


7/32

is prosocial, the greater is the image value of her prosocial act. However, the more people

think she is greedy, that is behaving prosocially to receive the extrinsic rewards, the lower

is the image value of her prosocial activity. A crucial property of image is its dependency

on visibility, after all image is a consequence of what others think. So if image value from aprosocial activity is positive, increasing the number of observers, increases the image value

of the prosocial activity and, as a consequence, leads to higher effort.

Even though we mainly focus on an activity with positive image value, following the same

logic increasing visibility should decrease effort for an activity with a negative image value.

This leads to the rst behavioral hypothesis on image-motivation:

Image-Motivation Hypothesis: Ceteris paribus, changing visibility changes the level of prosocial activity. For a positive (negative) image, increasing visibility increases (de-

creases) the level of prosocial activity.

Note that the case of self-image does not affect the Image-Motivation Hypothesis if self-

image is independent of visibility. In the experimental set-up, self-image can be thought of

as a constant in all conditions and does not affect the conclusions.

Introducing extrinsic incentives (or increasing thereof) affects prosocial behavior bothdirectly by increasing the extrinsic rewards, and through image motivation. On the one

hand, the relative price effect will increase prosocial activity, but, on the other hand, higher

personal benets associated with a prosocial activity tends to decrease its image value. The

intuition is that if an individual receives greater extrinsic rewards for the activity, she is a

suspect to behave prosocially for the extrinsic rewards rather than for intrinsic motivation.

Since being thought of as greedy is negative, this reduces the image value of the prosocial

activity. In such a case, offering a greater material reward may backre, depending on which

effect is stronger - higher extrinsic reward or lower image motivation.

Importantly, while the relative price effect is independent of visibility, the potential

crowding-out of image motivation crucially depends on visibility. As a consequence, the

the behavioral hypotheses based on this mode. Evidently, there are other possible setups one could consider.

7


8/32

effect of extrinsic rewards on image motivation implies that visibility may inuence the

effectiveness of material rewards. That is, if receiving an extrinsic reward reduces image

motivation (positive or negative), then greater publicity amplies this reduction and reduces

the effectiveness of material reward. This leads to our second hypothesis on the effectivenessof monetary incentives:

Effectiveness Hypothesis: Extrinsic rewards are less effective the greater is the visibility

of the prosocial act.

The Effectiveness Hypothesis implies crowding-out of image motivation. This conclusion

is possible even if extrinsic incentives do not have a negative net effect on prosocial activities,

i.e. when the direct increase in utility from extrinsic rewards outweighs any decline in

image motivation. Whenever prosocial activity level increases with monetary rewards it

is normally difficult to conclude whether extrinsic rewards crowd-out image motivation.

However, according to the Effectiveness Hypothesis, extrinsic rewards are less effective with

more visibility only as long as material rewards reduce image motivation. Therefore, the

case of higher prosocial activity with higher monetary rewards in which visibility reduces

the effectiveness of extrinsic rewards implies crowding-out. As a result, a detrimental effect

of extrinsic incentives (a negative net effect) is more likely to occur for a visible prosocial

effort than for a private one.

In the following sections, we test these hypotheses experimentally.

3 Experimental DesignClick for Charity

The experimental design has to fulll three criteria: (1) the prosocial task needs to have

clear image implications by allowing one to signal personal traits, (2) ones decision has to

be made either public or private, and (3) extrinsic monetary incentives have to be either

offered or not and to be publicly known.

8


9/32

In this study, participants could behave prosocially in a real-effort task: sequentially

pressing two keys X and Z on the keyboard for up to 5 minutes. For every completed

pair of clicks, we donated on the participants behalf to an assigned charity, hence, the name

Click for Charity. The donations were made according to a decreasing payment schedule:1 cent was donated for each of the rst 200 pairs of X-Z presses, 0.5 cents for each of the

next 200 pairs, 0.25 cents for each of the next 200 pairs, . . . , and 0.01 cents for each pair

above 1,200. Although marginally decreasing, the more pairs pressed, that is, the higher the

effort, the greater was the donation to the assigned charity. Before the actual task started,

each participant practiced the task for 30 seconds. 161 Princeton undergraduate students

participated in this study; none knew the nature of the experiment before entering the lab.

This study used a between-subjects 2x2x2 design. The conditions differed according to

the tasks visibility, the compensation scheme, and the nature of the charity, as explained

below.

The visibility of the task was controlled by randomly assigning subjects to make their

effort choice in public or in private. In the private condition, the effort decision was

made anonymously, while in the public condition, the choice was disclosed to others. At

the end of the task, we asked those in the public treatment to tell the other participants in

the lab which charity they were assigned to, whether they also earned money for themselves

(based on the payment scheme below), and how much money they donated to the charity

(and got for themselves).

The payment scheme was such that a random selection of participants could, in addition

to the donation, earn money for themselves. The private payment was exactly the same as

the donation schedule described above.The nature of the cause was manipulated by randomly assigning subjects to one of two

charities: the American Red Cross or the National Rie Association (NRA). Making efforts

to benet one of these charities is expected to have signaling value. Whether the value of

the signal is positive or negative depends at least partially on the norms and values of the

9


10/32

community and therefore on others (observers) opinions (as a result the two charities might

be perceived differently at Princeton and at a University in a pro-gun state). Therefore, we

classify the assigned charities as good or bad either not taking individual perceptions into

account (and relying on the consensus view) or by controlling for individuals perception of what others identify with. The rst classication analyzes the treatments with the Red Cross

and the NRA separately. According to our survey at Princeton, the majority of students

(i.e. students in the laboratory database) positively identies with the Red Cross (92% of

participants), and negatively identies with the NRA (72% of participants). According to

this consensus on campus, the Red Cross is clearly considered good and the NRA bad.

The second classication is based on the perceived majoritys view (on an 11-point scale

from -5 not at all identify to +5 very much identify) 4: a charity is considered good if

the reported perception is positive and bad if the reported perception is negative. 5 Seven

subjects stated a neutral perceived identication (exactly zero). When creating a dummy

for good or bad charities based on perceived others identication, we focused on the

remaining 154 subjects. However, when we control for perceived others perceptions using

the full scale or when we use the classication based on the consensus, all participants are

included. We will use both classications in the analysis.

The treatment with negative signaling value allows us to both examine how the results

for a good cause translates into a negative domain, and to refute alternative explanations

to the Image-Motivation Hypothesis. That is, one may think individuals put more effort in

public than in private not for the signal value of their actions, but for other reasons such as

showing they are very good in clicking. Such a tendency to show key pressing skills in public

should, of course, not interact with receiving an incentive and is, therefore not alternativeexplanation for the Effectiveness Hypothesis.

4In fact, we asked participants before the study started about the identication of Princeton studentswith a number of charities

5As one anonymous referee noted, false-consensus effect the tendency of people to think others aresimilar to them in views may be present. To minimize this effect, we selected extreme organizations: onethat the majority on campus does identify with, and one that the majority on campus does not identifywith. Hence, by construction the majority perceived the consensus correctly.

10


11/32

The following section presents the results of the study.

4 Results

We present the results in three steps: rst, we present the analysis for good and bad

causes separately, then we present the analysis for all causes together, and lastly we test

whether a physical limit on the ability to continuously press keys can explain the results.

Figure 1 shows clicking effort in public and in private, with and without private monetary

incentives for a good cause. Panel A shows effort for a good cause, dened according to

the consensus view at Princeton (the Red Cross). Panel B shows effort for a good cause

dened as what participants think others identify with. This classication takes into account

personal perception, while the classication of panel A does not. Comparing the two panels

we nd the same qualitative results.

First, without any private monetary incentives, subjects put forth signicantly more

effort in the public than in the private condition. In the public condition, they pressed, on

average, approximately 900 pairs for the Red Cross (822 pairs for a good charity), while

in the private condition, subjects pressed only about 517 for the Red Cross (548 pairs for agood cause). The difference is statistically signicant in a t-test 6 on the 1 percent level

and 5 percent level, respectively. In line with the Image-Motivation Hypothesis, subjects

exerted more effort for a good cause in public, where they are able to signal their effort to

many others.

[Figure 1 about here]

Second, while monetary incentives do not increase effort in public, they do increase effort

signicantly in the private condition. In the public condition, there is an insignicant decline

in effort from 900 to 814 key pairs on average for the Red Cross (from 822 to 702 key pairs

for a good charity). In the private condition, there is a signicant increase in effort from6We will use t-tests to compare all means, unless otherwise noted.

11


12/32

517 to 737 key pairs on average for the Red Cross ( p < 0.05) (from 548 to 740 key pairs on

average for a good cause, p < 0.05). Under both classications, the different impact of

incentives in public and in private is statistically signicant on the 5 percent level. 7 As we

show at the end of this section, this result is not driven by subjects in the public conditionreaching a physical limit. Therefore, this result supports the Effectiveness Hypothesis that

private monetary incentives are less effective in public than in private. This implies that

monetary incentives crowd-out the image motivation to behave prosocially.

Panel A and Panel B in Figure 2 show the result for the bad charity, dened according

to the consensus at Princeton (the NRA) or according to the perceptions of the participants,

respectively. For the bad charity case, without incentives, contributions are not signicantly

different in public and private. This result refutes alternative explanations arguing that par-

ticipants press more in public just because they are watched, independent of the image value

of their actions. One example of such explanation is that individuals press more in public to

show that they are good in pressing keys. If this were the case, we would nd signicantly

higher effort in the public sphere also for the bad charity. Although this result is appro-

priate to refute alternative explanations, according to the Image-Motivation Hypothesis we

would expect signicantly lower contributions in public relative to contributions in private

for the bad charity. Although this what we expected, we did not nd it in the data. Given

the low level of contribution for the bad in all conditions, we believe that what we see is

a censoring effet, a minimum level of key-pressing participants do, either because they feel

obliged to fulll their implicit agreement to participate, or because they are bored.

Regarding the Effectiveness Hypothesis, the gure show that monetary incentives increase

effort only in private. In public, monetary incentives slightly increase effort from 336 to 391for the NRA ( p = 0 .66) and slightly decrease effort from 308 to 260 for a bad cause

( p = 0 .76). However, in private, monetary incentives increase effort substantially. For the7The results are robust to focusing on participants who correctly perceived the consensus on campus

and therefore would not have been surprised if we had told them about the consensus at Princeton. SeeFigure A1 in the Appendix.

12


13/32

NRA, effort increases from 325 to 573 ( p = 0 .10) and for the bad cause from 204 to 559

key pairs ( p < 0.05). While the difference in effectiveness of incentives in private and public

is not statistically signicant for the NRA ( p = 0 .32), it is signicant on the 10 percent level

when classifying a cause as bad according to the perception of the participants ( p = 0 .06).In sum, offering monetary incentives is less effective at increasing effort in public than in

private even for a task with negative image value. As effort for a bad cause is at a very

low level, physical limit cannot explain the differential effect of monetary incentives in the

private and public conditions.

[Figure 2 about here]

Analyzing good and bad causes together, Table 1 shows the difference between private

and public conditions in an OLS regression where the number of key pairs is the dependent

variable. As independent variables, we control for the type of the assigned charity in two

ways: Panel A presents the analysis controlling for whether a participant was assigned to

the Red Cross or the NRA (a good or a bad cause according to the consensus view at

Princeton). Panel B shows the same analysis but controlling for the individual perception

of the majority view (on a full scale: -5 to +5). 8 The results are qualitatively the same in

both panels.

Column (1) in Table 1 shows that, in the private condition, monetary incentives increase

effort signicantly ( p < 0.01). In contrast, column (2) shows that for subjects in the public

sphere, monetary incentives do not increase individual effort. Both panels show that the

nature of the cause is important for peoples effort, but the cause seems to be more important

in public than in private. This is consistent with the view that the signal matters more inpublic than in private. As seen in column (3), the difference in the effectiveness of monetary

incentives in the public versus the private conditions (i.e. the interaction term of private

monetary incentives and the public condition) is substantial and statistically signicant either8The results are robust to controlling for a dummy for good and bad charity based on perceived

identication of others. See table A1 in the Appendix.

13


14/32

close to the 95 percent level ( p = 0 .059) in Panel A or at the 99 percent level ( p < 0.01) in

Panel B. Also, column (3) shows that the nature of the cause is signicantly more important

in public than in private. The results support both the Image-Motivation Hypothesis, that

the signal matters more in public, and the Effectiveness Hypothesis, that monetary incentivesare less effective in the presence of image motivation, i.e. if an activity with image reward

is public.

[Table 1 about here]

One concern while interpreting the differential effect of monetary incentives in the public

compared to the private condition is that the effect may be due to a physical limit. Subjects

exerting effort for a good cause in the public condition may reach a physical limit even

without incentives. In this case, it might be impossible to increase effort with monetary

incentives, even if subjects wanted to. While the analysis for the bad cause suggests

otherwise, since effort is at a low level in this condition, we nevertheless tested explicitly

whether participants reached a physical limit in the good-cause condition. To nd where

peoples physical limit is, we ran an additional control treatment, the Limit treatment, in

which 26 students from the same subject pool at Princeton participated. In this treatment,participants performed the same key-pressing task for 5 minutes, and their effort decision

remained anonymous. However, in this condition the task was not a donation task; as the

subjects could only earn money for themselves by pressing the keys, and they were paid high

nancial incentives to do so. Subjects got a at rate of 0.11 cents per pair, plus a bonus,

which increased from $2, if they pressed more than 1,000 pairs, to $20, if they pressed more

than 1,500 pairs.

The average number of paired key presses in the Limit treatment was 1,290 (s.e. 44) pairs,

while subjects pressed only 857 (s.e. 61) key pairs for the Red Cross in the public sphere. The

difference between the key presses in the two conditions is statistically signicant ( p < 0.01).

Analyzing the distribution of key pairs for the Red Cross in public and in the Limit treatment,

we nd that while in the Limit treatment nobody pressed less than 900 pairs, 50 percent

14


15/32

of the participants did so for the Red Cross (see Figure A2 in the Appendix). Therefore,

the ineffectiveness of monetary incentives in the public condition cannot be attributed to a

physical limit, as most subjects could have increased their effort substantially.

5 Field StudyBike for Charity

The Click for Charity results support the view that extrinsic incentives may have detrimental

effects by crowding-out image motivation. This depends on the extent of public exposure

of the prosocial activity, which, in turn, has important implications for the use of extrinsic

incentives in facilitating contributions to public goods. To strengthen the policy implications

of these results, below we present a rst-pass illustration of the results in the eld.

To illustrate whether the effect found in the Click for Charity experiment also translates

to a eld setting, we launched a similarly-designed campaign at the MIT gym, titled Bike

for Charity. In the Bike for Charity campaign, participants could donate to a good or

bad charity by cycling on stationary bikes for up to 10 minutes, at a donation rate of $1

per mile. As in Click for Charity we implemented a between-subject 2x2x2 design. The

visibility was manipulated by randomly assigning subjects to bike in the public view (on thesecond oor of the gym, where everyone works out) or in a private location (the third oor

of the gym, in a private room). The participants were also assigned to receive either private

monetary rewards for themselves or not. Subjects who participated in the rst few weeks of

the campaign did not receive any private monetary incentives, while those who participated

in the last weeks of the campaign, did. The cause of the charity was randomly manipulated

as in Click for Charity. 9

Due to the lack of a randomized compensation scheme, and to the fact that participants

knew in advance that they would be randomly assigned to bike for either a good or a9In the Bike for Charity we used four controversial charities: Two of the four focus on gun control and

are either pro or against stronger gun control (the National Rie Association (NRA) and the EducationalFund to Stop Gun Violence), the other two charities focus on abortion and lobby either for or against it (thePro-Life Association and the Pro-Choice Action League).

15


16/32

bad charity, selection bias is a concern. More specically, we expect that individuals

whose image is affected the most by monetary incentives will avoid conditions with private

monetary incentives. This, in turn, could dampen the interaction between extrinsic incentives

and image concerns. Similarly, if the nature of the cause is unclear (the most extreme caseis random assignment, as in our study), the more image-concerned a person is, the less likely

she would be to participate in such a campaignto avoid the chance of being associated with

a bad cause. Both selection biases act against the image and the crowding-out hypotheses.

Despite the limits of the Bike for Charity design, the results present suggestive eld ev-

idence for the Effectiveness Hypothesis. 151 participants were assigned to bike either for

a good or a bad charity, where good and bad charities are dened according to

whether participants perceive MIT students identication as positive or negative, respec-

tively. For these participants, adding private incentives is, on average, more effective in

private than in public. Table 2 shows this effect in an OLS regression with distance biked

in miles as the dependent variable. We use the same independent variables as in Click for

Charity, plus a variable for the participants gender. Table 2 shows that extrinsic incentives

increases effort in the private condition (Column 1; p < 0.01) but not in the public condition

(Column 2; p = 0 .96), and as can be seen in column 3 this difference is statistically signicant

( p = 0 .057).10

[Table 2 about here]

The effect obtained in the eld is not as strong as in the lab. However, as discussed, the

lab results are obtained in a fully randomized setting where participants could not self-select

into a condition based on its characteristics (a good/bad cause, and with/without privatemonetary incentives), while in the eld they partly could. Therefore, the suggestive evidence

is encouragingin spite of the selection bias we observe differential effects of monetary10 Adding individuals to the analysis who believe that the charities are seen by other students as neither

positive nor negative and therefore stated a 0 on scale from -5 not at all identify to +5 very muchidentify, as in Click for Charity, does not change the results qualitatively but increases the standard errorsof the estimations.

16


17/32

incentives between the public and the private conditions. Future research should further

investigate the interaction between extrinsic incentives and image motivation in the eld.

6 Conclusions

The results of the laboratory experiment, illustrated also in a eld study, support our hy-

pothesis that monetary incentives depend crucially on visibility: monetary incentives are

more effective in facilitating private, rather than public, prosocial activity. People want to

be seen as doing good ; without extrinsic incentives, an observer will attribute the prosocial

act to ones good trait which motivates people to behave prosocially. But with extrinsic

incentives, the signal of a prosocial act gets diluted as one might behave prosocially mainly

to do well . The image value decreases, and the incentive becomes less effective. If no one

is watching (i.e. the prosocial decision is private) the incentive to be doing well cannot

dilute any signal to others, and consequently extrinsic incentives are very likely to increase

prosocial behavior.

Our results have important policy implications; for example, if a government considers a

tax benet policy to facilitate the adoption of a new environmentally friendly technology, itshould expect the policy to be more successful for a non-visible technology, such as environ-

mentally friendly water boilers, relative to a visible technology, such as hybrid cars. This is

because hybrid cars, most of which are clearly visible, may partly be purchased as a signal-

ing medium, while water boilers probably are not. By giving tax benets, the government

might unintentionally damage the signaling value vested in hybrid cars. The net effect of the

incentive depends on the strength of the price effect compared with the crowding-out effect.

It is important to note that image crowding-out by monetary incentives is shown here

when both the prosocial decision and the extrinsic incentives are public. If extrinsic incentives

exist but are not publicly known, others opinion should be the same as if no extrinsic rewards

existed. Consequently, major donors to cultural organizations might not want to make their

17


18/32

extrinsic incentives (e.g. favorable seats) public, as it would dilute the image signal of

their contribution. If decision-makers, such as fund-raisers or policy makers, anticipate the

Effectiveness and Crowding-Out Hypotheses, they should use fewer public extrinsic incentives

for visible prosocial activities. This brings us back to Titmuss and his intuition that monetaryincentives might reduce pro-social behavior. The research in this paper contributes to the

discussion in showing that Titmusss claim crucially depends on the visibility of the prosocial

decision. Therefore if blood donations are sufficiently public, there is good reason to doubt

the effect of extrinsic incentives and even to expect detrimental effect of monetary incentives.

In this study we show empirically that image motivation and visibility are important

for prosocial behavior. However, there are several open questions which this study did not

answer and are left for future research. One is how the conclusion of this research translates

to other circumstances. In particular, there are societies where volunteering may not be

perceived as honorable as in the US. In these cases image might not be a signicant drive in

behaving prosocially. This leads to the second point: how is image motivation constructed?

Norms, perception and personal views are clearly its ingeredients, but we dont know the

mechanism by which image is formed. In this study we overcome this issue by selecting

extreme charities where norms perceptions and personal views are aligned. However, it

would be interesting to understand what would happen in cases where private views may

not be alligned with others views or when the majority view is not clear. Last, but not least:

does crowding-out of image motivation depend on the type of the extrinsic incentive used?

In this study we explore the effect of monetary incentives, which may be different from the

effect of non-monetary incentives. To continue with the hybrid car example, the incentive of

using the carpool lane may not dilute the signal as much as monetary incentives would.

18


19/32

Figure 1: Effect of Incentives on Prosocial Behavior

Panel A: Red Cross Panel B: GoodCause

0

200

400

600

800

1000

1200

Wi thout Incentive With Inc enti ve

Payment Scheme

N o

. o

f K

e y p r e s s

Private Public

0

200

400

600

800

1000

1200

Without Incentive With Incentive

Payment Scheme

N o

. o

f K

e y p r e s s

Private Public

Note : Error bars are standard errors of the mean. Panel A shows effort for the Red Cross (the majorityat Princeton positively identies with this charity). Panel B shows effort for a good cause according toparticipants perception of others identication at Princeton.

19


20/32

Figure 2: Effect of Incentives on Effort for a Bad Cause

Panel A: NRA Panel B: BadCause

0

200

400

600

800

1000

1200

Witho ut Incentive Wi th Incentive

Payment Scheme

N o

. o

f K

e y p r e s s

Private Public

0

200

400

600

800

1000

1200

Wi thout Incentive With Inc enti ve

Payment Scheme

N o

. o

f K

e y p r e s s

Private Public

Note : Error bars are standard errors of the mean. Panel A shows effort for the NRA (the majority at Princetondoes not identies with this charity). Panel B shows effort for a bad cause according to participants perceptionof others identication at Princeton.

20


21/32

Table 1: Effect of Private Incentives in Click for Charity

(1) (2) (3)Private Public All

Panel A: Controlling for Red Cross and NRAPrivate monetary incentive (=1) 231.95 0.64 231.95

(2.82)*** (0.01) (2.81)***Public sphere (=1) 26.94

(0.25)Private incentive*Public -231.31

(-1.91)*Nature of cause (RC=1) 179.34 498.02 179.34

(2.08)** (5.69)*** (2.08)**Nature of Cause*Public 318.68(2.59)***

Constant 332.05 358.99 332.05(4.27)*** (4.78)*** (4.27)***

N 82 79 161R squared 0.15 0.28 0.22

Panel B: Controlling for perceived identication of othersPrivate monetary incentive (=1) 240.80 -88.88 240.80

(3.01)*** (-0.95) (3.01)***

Public sphere (=1) 152.45(1.80)*

Private incentive*Public -329.67(-2.68)***

Perceived Identication of others 43.55 89.35 43.55(3.00)*** (5.52)*** (3.00)**

Perceived Identication*Public 45.80(2.11)**

Constant 384.43 536.88 384.43(7.11)*** (8.25)*** (7.11)***

N 82 79 161R squared 0.20 0.29 0.25

Note : Dependent variable: Number of key pairs. OLS regressions. t -values inparenthesis. Robust standard errors.Level of signicance : * p < 0.1, ** p < 0.05, *** p < 0.01

21


22/32

Table 2: Effect of Private Incentives in Bike for Charity


Private monetary incentive (=1) 0.56 -0.02 0.56

(2.78)*** (-0.07) (2.81)***Public sphere (=1) 0.28(1.34)

Private incentive*Public -0.58(-1.93)*

Good cause (=1) 0.84 0.84 0.85(3.75)*** (3.21)*** (4.84)***

Female (=1) -0.54 -0.46 -0.50(-2.71)*** (-2.08)** (-3.30)***

Constant 1.50 1.75 1.48(6.27)*** (6.04)*** (7.30)***

N 69 82 151R squared 0.33 0.18 0.24

Note : Dependent variable: Distance in miles. OLS regressions. t -values inparenthesis. Robust standard errors. Charity classied according to participantsperception of whether MIT students identify with charity (Good if othersidentication is greater than zero and Bad if others identication is lowerthat zero on a scale from -5 not at all identify to +5 very much identify).Level of signicance : * p < 0.1, ** p < 0.05, *** p < 0.01

22


23/32

ReferencesAkerlof, George A. , A Theory of Social Custom, of Which Unemployment May Be One Con-

sequence, Quarterly Journal of Economics , 1980, 94 (4), 74975.

Andreoni, James and B. Douglas Bernheim , Social Image and the 50-50 Norm, Mimeo ,2007.

and Ragan Petrie , Public Goods Experiments without Condentiality: A Glimpse intoFund-Raising, Journal of Public Economics , 2004, 88 (7-8), 160523.

Benabou, Roland and Jean Tirole , Incentives and Prosocial behavior, American EconomicReview , 2006, 96 (5), 16521678.

Blumkin, Tomer and Efraim Sadka , A Case for Taxing Charitable Donation, Journal of Public Economics , 2007, 91 (7-8), 15551564.

Dana, Jason, Daylian M. Cain, and Robyn M. Dawes , What You Dont Know Wont Hurt

Me: Costly (but Quiet) Exit in Dictator Games, Organizational Behavior and Human Decision Processes , 2006, 100 (2), 193201.

Deci, Edward L. , Intrinsic Motivation , New York: Plenum Press, 1975.

Ellingsen, Tore and Magnus Johannesson , Generosity, Mimeo , 2007.

and , Pride and Prejudice: The Human Side of Incentive Theory, American EconomicReview , 2008, p. Forthcoming.

Falk, Armin and Michael Kosfeld , The Hidden Cost of Control, American Economic Review ,2006, 96 (5), 161130.

Fehr, Ernst and Armin Falk , Psychological Foundations of Incentives, European EconomicReview , 2002, 46 , 287324.

and John A. List , The Hidden Costs and Returns of Incentives - Trust and Trustworthinessamong CEOs, Journal of the European Economic Association , 2004, 2 (5), 74371.

and Klaus Schmidt , Theories of Fairness and Reciprocity - Evidence and Economic Applica-tion, in M. Dewatripont, L. P. Hansen, and S. J. Turnovsky, eds., Advances in Economics and Econometrics - 8th World Congress, Econometric Society Monographs , Cambridge: CambridgeUniversity Press, 2003, pp. 20857.

Frey, Bruno S. , Not Just for the Money. An Economic Theory of Personal Motivation , Chel-

tenham, UK and Brookeld, USA: Edward Elgar, 1997.and Felix Oberholzer-Gee , The Cost of Price Incentives: An Empirical Analysis of Moti-

vation Crowding-Out, American Economic Review , 1997, 87 (4), 74655.

Glazer, Amihai and Kai A. Konrad , A Signaling Explanation of Charity, American Eco-nomic Review , 1996, 86 (4), 101928.

Gneezy, Uri and Aldo Rustichini , A Fine Is a Price, Journal of Legal Studies , 2000, 29 (1),118.

23


24/32

and , Pay Enough or Dont Pay at All, Quarterly Journal of Economics , 2000, 115 (3),791810.

Harbaugh, William T. , The Prestige Motive for Making Charitable Transfers, American Eco-nomic Review , 1998, 88 (2), 27782.

, What Do Donations Buy? A Model of Philanthropy Based on Prestige and Warm Glow,Journal of Public Economics , 1998, 67 , 269284.

Heyman, James and Dan Ariely , Effort for Payment: A Tale of Two Markets, Psychological Science , 2004, 15 (11), 78793.

Hollander, Heinz , A Social Exchange Approach to Voluntary Cooperation, American EconomicReview , 1990, 80 (5), 115767.

Meier, Stephan , A Survey on Economic Theories and Field Evidence on Pro-Social Behavior,in Bruno S. Frey and Alois Stutzer, eds., Economics and Psychology: A Promising New Cross-Disciplinary Field , Cambridge: MIT Press, 2007, pp. 5188.

Mellstroem, Carl and Magnus Johannesson , Crowding out in Blood Donation: Was TitmussRight?, Journal of the European Economic Association , 2008, p. Forthcoming.

Rege, Mari and Kjetil Telle , The Impact of Social Approval and Framing on Cooperation inPublic Good Situations, Journal of Public Economics , 2004, 88 (7-8), 162544.

Sector, The Independent , Giving and Volunteering in the United States 2001, TechnicalReport, The Independent Sector 2001.

Soetevent, Adriaan R. , Anonymity in Giving in a Natural Context - an Economic Field Ex-periment in 30 Churches, Journal of Public Economics , 2005, 89 (11-12), 230123.

Titmuss, Richard M. , The Gift Relationship , London: Allen and Unwin, 1970.

24


25/32

A Appendix (Web Appendix)

A.1 Hypothesis Derived from Benabou and Tirole (2006)In Benabou and Tirole (2006) an individuals incentives to behave prosocially are divided

into three basic motivations: intrinsic ( va ), extrinsic ( yvy ), and image motivation R(a, y).Formally, an agents utility is:

U (a) = ( va + yvy )a + R(a, y) C (a),

where a is the extent to which the agent behaves prosocially, y is the extrinsic or monetaryrewards (valued at vy )11 , and C (a) is the cost of behaving prosocially. ( va , vy ) are assumedto be normally distributed, C (a) = ka

2

2 , and R(a, y) is dened as:

R(a, y) = x[ a E (va |a, y) y E (vy |a, y)].

Here, x is the visibility of action a, a is the agents level of concern for being consideredan altruistic or prosocial human-being, E (va |a, y ), and y is the agents level of concern for

being perceived as greedy, E (vy |a, y).

In the analysis leading to the behavioral hypotheses we focus on the case where types

v (va , vy) are distributed normally in the population ( va , vy ) N vavy

, 2a ay

ay 2y,

va , vy > 0, and individuals are assumed to have constant image attitudes ( a , y)

(x a , x y) = ( a , y). Under these assumptions, the optimal level of a is given by Proposition

1 in BT:

Proposition 1 (BT1) Let all agents have the same image concern (a , y ), and cost func-tion c(a) = ka 2/ 2. There is a unique (differentiable-reputation) equilibrium, in which an agent with preferences (va , vy) contributes at the level

a=va + vyy

k+ a (y) y (y),

=va + vyy

k+ xr (y)

where (y) 2a + y ay

2a +2 y ay + y2 2y

and y (y) 1 (y). The reputational returns are E (va |a ,y )

a =

(y)k and E (vy |a ,y )

a = (y)k, resulting in a net value kxr (y) = k(a (y) y (y)), or r (y) = R (a,y )a |x =1 ,independent of a.

11 y represents personal extrinsic rewards only. It does not capture resources available to a charity ororganization which benets from ones prosocial action a . If the individual cares for the total resources forthe organization, this may be incorporated into the intrinsic motivation, va .

25


26/32

Proof. The proof of this proposition is omitted and can be found in BT.

The aggregate supply, a, is:

a =va + vy y

k+ xr (y),

where (va , vy ) are the populations averages, and xr (y) is the constant (marginal) image

motivation 12 .

A straight-forward conclusion is the Image-Motivation Hypothesis, next:

Proposition 2 a (y)x > 0 r (y) > 0.

The image motivation hypothesis is stated below:

Image-Motivation Hypothesis: Ceteris paribus, changing visibility x changes prosocial

activity a. For a positive image (r (y) > 0), increasing x increases a. For a negative

image (r (y) < 0), increasing x decreases a.

Further calculation leads to the following proposition:

Proposition 3 Let ay = 0 and dene y a . If r (y) > 0,a (y)

x < 0.

Proof. The change in prosocial behavior as we change the extrinsic reward y is given by:

a (y) =vyk

+ a (y) y (y).

a (y)x

= a (y) y (y)

(y) =2a + yay

2a + 2 yay + y22y=

2a2a + y22y

if ay = 0 (y) = 2 (y)(y)

(y) =y2y + ay

2a + 2 yay + y

2

2y

=y2y

2a + y

2

2y

if ay = 0 (y) =2y

2a + y

2

2y

((y) y (y))

a (y)x

= a (y) y (y)

= (y) a (y) y (y) a (y)(y) y (y)(y)

y12 r (y ) = R (a,y )a |x =1 , independent of a.

26


27/32

If r (y) > 0, i.e., a (y) y (y) > 0,a (y)

x < 0

Proposition 3 implies that, provided a positive net image when x = 1 (private) , the

greater the publicity of ones actions, the less effective are the extrinsic rewards associated

with giving. Note that a (y)x < 0 r (y) = a (y) y (y) < 0. This leads us to the

Effectiveness Hypothesis, and its implication for crowding-out, stated and explained below:

Effectiveness Hypothesis: Extrinsic rewards, y, are less effective the greater is the visi-

bility of the prosocial act, x.

The Effectiveness Hypothesis implies crowding-out of image motivation, r (y) < 013 . Thisconclusion is possible even if extrinsic incentives do not backre, i.e., a (y) > 0. This occurswhenever the direct increase in utility from extrinsic rewards outweighs any decline in imagemotivation . As a result, we have three possible cases:

1. a (y) 0 : this clearly indicates crowding-out

2. a (y) > 0, and r (y) < 0 : crowding-out of image motivation

3. a (y) > 0, and r (y) 0 : there is no crowding-out.

Whenever a (y) > 0, it is difficult to conclude whether extrinsic rewards crowd-out image

motivation. However, according to the Effectiveness Hypothesis, extrinsic rewards are lesseffective with more visibility only as long as material rewards reduce image motivation.

Therefore, the case of a (y) > 0 in which visibility reduces the effectiveness of extrinsic

rewards implies crowding-out. As a result, a detrimental effect of extrinsic incentives is more

likely to occur for a visible prosocial effort than for a private one.

A.1.1 Robustness to other specications

The hypotheses above are for the simple case of no correlation between altruistic taste andpersonal value of extrinsic rewards ( ay = 0) . However, this does not affect our hypotheses.The Image-Motivation Hypothesis says that if image motivation is indeed one of the driversof prosocial behavior and it is positive for some visibility level, then increasing the visibility

13 This also holds for the case of heterogenous image concerns provided that the direct effect of visibilityon the effectiveness of monetary rewards, r (y ), dominates the indirect effect of visibility via image variance(one may suspect prosocial behavior for image concerns).

27


28/32

of ones actions will increase the level of prosocial behavior a. This hypothesis holds equallywell even if ay = 0 . To see this, note that the image motivation, r (y), is inuenced by ay :

r (y) = a (y) y (y)

= a (2a + yay ) y(ay + y2y )

(2a + 2 yay + y22y )

However if r (y) is positive, increasing the visibility of ones action by x > 1 acts only as an

amplier: xr (y).Similarly, the claim made in the Effectiveness Hypothesis is valid provided that ay is

not too large. To see this, lets examine the following object:

a (y)x

= ax

(y) yx

(y).

Since (y) and (y) are functions of ay , relaxing the assumption that ay = 0 does inuence

the Effectiveness Hypothesis. However, it can be shown that, as long as ay is not too large,

the conclusion still holds.Proof.

(y) =2a + yay

2a + 2 yay + y22y (y) =

2a ay 22a y2y y22y ay2a + 2 yay + y22y

2

(y) =y2y + ay

2a + 2 yay + y22y (y) =

2y 2a y24y 22ay 2y2y ay2a + 2 yay + y22y

2

a (y)x

= a (y) y (y)

= a 2a ay 22a y2y y22y ay

2a + 2 yay + y22y2 y

2y 2a y24y 22ay 2y2y ay2a + 2 yay + y22y

2

If ay = 0 :

a (y)

x=

2ay (2 y) + ay 2 yy2y a 2a a y22y 2 a 2a y2y y 2y 2a y24y

[2a + 2 yay + y

2

2y ]

2

a (y)

x< 0

(2 y )2ay + ay (2y y 2y a 2a a y22y ) a 2y2a 2y y [2a 2y y24y ] < 0

(2 y )2ay + ay (2y y 2y a 2a a y22y ) a 2y2a 2y y[2a 2y y24y ] = 0

28


29/32

(A)x2 + x(B) (C ) = 0 .

For ay = 0 a (y)x < 0; this implies that there is a solution to the above quadratic equation

in ay (C < 0 B 2 4AC > B 2 > 0). Therefore, as long as ay is small enough, a (y)x < 0.

29


30/32

A.2 Appendix Figures

30


31/32

Figure A1: Effect of Incentives for Red Cross and NRA: Restricted to Participants whoCorrectly Perceive Majority Identication

Panel A: Red Cross Panel B: NRA

0

200

400

600

800

1000

1200

Wi thout Incenti ve With Incentive

Payment Scheme

N o

. o

f K

e y p r e s s

Private Public

0

200

400

600

800

1000

1200

Witho ut Incentive With Incentive

Payment Scheme

N o

. o

f K

e y p r e s s

Private Public

Note : Error bars are standard errors of the mean. Panel A shows effort for the Red Cross and Panel B for theNRA (the majority view at Princeton) for individuals who correctly perceive the majority view.

Figure A2: Cumulative Distribution of No. of Key Pairs

0

10

20

30

40

50

60

70

80

90

100

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

No. of Key pairs / 100

C u m

u l a t i v e

D i s t r i b u

t i o n

Click for Red Cross Limit' treatment

31


32/32

Table A1: Effect of Private Incentives in Click for Charity


Panel APrivate monetary incentive (=1) 252.97 -90.81 252.97

(3.01)*** (-0.98) (3.01)***Public sphere (=1) 72.89


(-2.75)***Good cause (=1) 269.68 483.51 269.68

(3.00)*** (5.08)*** (3.00)***Nature of Cause*Public 213.82

(1.63)

Constant 249.29 322.18 249.29(3.29)*** (3.82)*** (3.29)***N 77 77 154R squared 0.21 0.25 0.24

Panel BPrivate monetary incentive (=1) 239.03 -88.52 239.03

(2.93)*** (-0.97) (2.93)***Public sphere (=1) 15.36


(-2.68)***Good Cause 216.19 486.74 216.19

(2.51)** (5.24)*** (2.51)**Good Cause*Public 270.55

(2.13)**Constant 302.29 317.65 302.29

(3.99)*** (4.02)*** (3.98)***N 82 79 161R squared 0.17 0.26 0.22

Note : Dependent variable: Number of key pairs. OLS regressions. t -values in

parenthesis. Robust standard errors. Charity classied according to partici-pants perception of whether Princeton students identify with charity (Goodif others identication is greater than zero and Bad if others identication islower that zero on a scale from -5 not at all identify to +5 very much iden-tify). Panel A: Those stating 0 are excluded from the analysis. Panel B: Forthose stating 0, the assigned charity got classied according to the perceptionof the majority: NRA is perceived as a bad cause and Red Cross as a goodcause.Level of signicance : * p < 0.1, ** p < 0.05, *** p < 0.01

doing good or doing well? image motivation

Documents