altruism identity and financial returns...

48
1 Altruism, Identity and Financial Return: An Experiment on Microfinance Lending 1 This version: February 21, 2015 Josie I Chen and Louis Putterman Abstract. We design a laboratory experiment to study willingness to lend and preference over borrowers in one-to-one on-line micro-finance lending. We find that both financial return and intrinsic (philanthropic) motivation affect the amount lent, with no evidence that the former crowds out the latter. Ethnic and other forms of identity have long been found to play a part in charitable and trust game contributions, but past studies have not ruled out impacts of face-to-face social pressures and differential assessments of riskiness. Our design eliminates social pressure and lets us control for two factors we elicit from the lenders/subjects: assessments of the likelihood of repayment by borrowers, as well as perception of their deservingness. We find that lenders are willing to trade greater risk in order to help more needy borrowers, but at a rate sensitive to the financial return condition. We find that lenders still show a statistically significant preference for borrowers with whom they share gender and ethnic similarity even after controlling for perceived riskiness, neediness, and other factors including physical attractiveness and weight. Keywords: microfinance, social entrepreneur, pro-social preferences, identity, gender, ethnicity JEL codes: C91, D64, J15, J16 1 We thank Pedro Dal Bó, Andrew Foster, Mark Dean and seminar participants at Brown’s Theory Workshop and at 2014 Economic Science Association International Meetings for helpful comments. This research was supported in part by Russell Sage Foundation and by William R. Rhodes Center.

Upload: others

Post on 20-Sep-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  1  

Altruism, Identity and Financial Return:

An Experiment on Microfinance Lending1

This version: February 21, 2015

Josie I Chen and Louis Putterman

Abstract. We design a laboratory experiment to study willingness to lend and preference over borrowers in one-to-one on-line micro-finance lending. We find that both financial return and intrinsic (philanthropic) motivation affect the amount lent, with no evidence that the former crowds out the latter. Ethnic and other forms of identity have long been found to play a part in charitable and trust game contributions, but past studies have not ruled out impacts of face-to-face social pressures and differential assessments of riskiness. Our design eliminates social pressure and lets us control for two factors we elicit from the lenders/subjects: assessments of the likelihood of repayment by borrowers, as well as perception of their deservingness. We find that lenders are willing to trade greater risk in order to help more needy borrowers, but at a rate sensitive to the financial return condition. We find that lenders still show a statistically significant preference for borrowers with whom they share gender and ethnic similarity even after controlling for perceived riskiness, neediness, and other factors including physical attractiveness and weight.

Keywords: microfinance, social entrepreneur, pro-social preferences, identity, gender, ethnicity

JEL codes: C91, D64, J15, J16

                                                                                                               1  We thank Pedro Dal Bó, Andrew Foster, Mark Dean and seminar participants at Brown’s Theory Workshop and at 2014 Economic Science Association International Meetings for helpful comments. This research was supported in part by Russell Sage Foundation and by William R. Rhodes Center.

Page 2: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  2  

1. Introduction

After Mohammad Yunus and the Grameen Bank were awarded the Nobel Peace

Prize in 2006, innovation in microfinance has been progressing and providers of

financial services to the poor continue to evolve. Recently, social entrepreneurs have

combined microfinance lending and on-line crowd-funding investment allowing for

increased accessibility. On average, the microfinance borrowers pay 30% interest, most

of which goes to the financial intermediaries to cover their operating expenses, rather

than to the lenders.2 It is difficult to rule out that substantial amounts of surplus cash

might be obtainable from savers in developed countries such as the United States --

where interest rates on savings and checking accounts are quite low -- if a small fraction

of the interest paid by microfinance borrowers could be shared with them. At least

hypothetically, such sharing of interest between intermediaries and individual online

lenders could benefit both the lenders and the severely capital-constrained micro-

entrepreneurs in less-developed countries. However, most intermediaries steer clear of

offering interest for fear of destroying the philanthropic appeal of microfinance lending.

What we see now is that the scale of existing rich country to poor country lending via

microfinance intermediaries remains small.3 4

                                                                                                               2 See Rosenberg, Richard, et al. "Microcredit interest rates and their determinants: 2004–2011." Microfinance 3.0. Springer Berlin Heidelberg, 2013. 69-104. 3 Kiva.org is perhaps the largest microfinance websites in the world, with about $0.6 billion lent since its founding in 2005. Lenders at kiva.org recoup repaid principal but do not receive interest. There are many microfinance platforms differing in some substantial ways, including the amount of information given about the borrowers, whether lenders receive the principal and whether lenders earn interest. For example, lenders at worldvision.org select the initial loan recipient, but when the loan is repaid, World Vision keeps the cash and recycles it over and over again to the other entrepreneurs in the same country- similar to the wokai.org offering in China. Lenders at microplace.org select different "loan groups" with specific interest, but without knowing which specific borrowers made up these loan groups. When the loan group repays the loan, the lenders receive the loan amount and also the interest. Lenders at zidisha.org, select the loan recipient and receive the principal and interest when a loan is due. 4 According to the World Bank, there are approximately 2.5 billion people who do not have bank accounts and whose business could benefit from microcredit. However, data reported at the 2014 Microcredit

Page 3: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  3  

Motivated by the idea that microfinance lending might be falling short of its

potential for scaling up, we study how intrinsic and extrinsic incentives influence the

willingness of rich country residents to lend to small entrepreneurs in developing

countries.5 A few studies have shown that existing microfinance lenders were motivated

mostly by intrinsic incentives (Choo et al., 2014; Liu et al., 2012). Specifically, Chen et

al. (2014) has explored how intrinsic incentives such as team competition could boost

peoples’ willingness to lend. A large amount of research into donation, volunteering,

and public goods shows that increasing extrinsic incentive may decrease peoples’

willingess to “do good things” (i.e. things with a positive externality), because the

intrinsic incentives are diminished by the existence or strength of extrinsic incentives.

The evidence of this “crowd-out effect” is shown in the empirical data, natural field

experiments, and lab experiments (Ariely, Bracha and Meier, 2009; Deci, 1971; Frey,

1994; Frey and Oberholzer-Gee, 1997; Gneezy, Meier and Rey-Biel, 2011; Greezy and

Rustichini, 2000; surveys in Lane, 1991; surveys in Ryan and Deci, 1985; and Titmuss,

1971).

These studies, however, vary across just one of the dimensions, either intrinsic or

extrinsic. It is unknown how these two incentives interact. For example, given an

extrinsic incentive, does addition of an intrinsic incentive boost peoples’ willingness to

do good things? If so, is the marginal effect of the intrinsic incentive increasing or

decreasing in the strength of the extrinsic one? To explore how the dimensions of

financial inducement and pro-social motives interact in motivating micro-finance

                                                                                                                                                                                                                                                                                                                                         Summit in Mérida, Mexico shows that only 116 million of the world’s poorest are reached by microfinance as of December 30, 2012. 5 We note, as an aside, that the extent of the benefits and possible harms caused by micro-finance are debated (see, for example, Morduch, 2013). That discussion must remain beyond the scope of our paper.

Page 4: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  4  

lending, we vary these two dimensions across treatments. In the first stage of our

experiment, we ask subjects to choose a loan amount. We vary the first dimension of

whether or not borrowers have a financial stake in their loan’s repayment and how

profitable the lending could be. On the other hand, to shape the perception of neediness

and the intrinsic incentives, we vary the second dimension of whether or not the initial

description of the lending opportunity emphasizes its potential philanthropic appeal.

We find that subjects are motivated to lend to microfinance borrowers both by

financial inducement and by pro-social motives. While financial incentives matter, their

importance decreases when lenders know that borrowers are micro-entrepreneurs from

poor developing countries. While subjects who are informed that borrowers are micro-

entrepreneurs lend more than subjects who are not informed, the difference is significant

only when there is no financial stake for subjects.

We explore the effect of financial returns not only upon lenders’ willingness to

lend but also on their preference over borrowers. Research has shown that given

different financial return conditions, lenders’ choices of borrowers are different:

conditional on no possibility of repayment, which makes the lending decision more like

a donation, research suggests that donors choose borrowers from a more needy, socially

preferred, or morally preferred group in order to maximize the social impact (Eckel and

Grossman, 1996; Fong and Oberholzer-Gee, 2011). Conditional on a given positive

interest rate, standard theory suggests that lenders will choose a borrower perceived as

having lower default risk (henceforth, “a less risky borrower”) in order to maximize the

expected payoff. It is not currently known whether lenders still prefer less risky

Page 5: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  5  

borrowers in the former case and whether lenders still prefer more needy borrowers in

the latter case.

After eliciting subjects’ willingness to lend, influenced by rate of return and

philanthropic or neutral description of lending opportunity, we ask them to report their

assessments of the likelihood of repayment by each of twelve specific borrowers, as well

as those borrowers’ individual levels of neediness. We then have subjects rank order

their three most preferred loan recipients. We find that there is a trade-off between

borrowers’ neediness and riskiness, at a rate sensitive to the financial return condition.

When there is a financial stake, lenders’ preferred ranking of borrowers responds

equally positively to a 1 standard deviation increase in perceived neediness or a 2

standard deviation decrease in perceived riskiness level. When there is no financial

stake, the impact on the ranking of a 1 standard deviation increase in perceived

neediness level is roughly equal to that of a 1 standard deviation decrease in perceived

riskiness level.

Our second stage structure also allows us to assess whether gender or ethnic ties

influence microfinance lending. Ethnic and other forms of identity have long been

thought to play an important role in private and non-profit contributions. Recent research

has documented the importance of such factors in charitable and microfinance donation

as well as in dictator game and trust game contributions (Fershtman and Gneezy, 2001;

Fong and Luttmer; 2009; Galak, Small, and Stephen, 2013; Glaeser et. al, 2000;

Rotemberg, 2012). A reason contributors may have shown a preference towards those

sharing the same ethnic or racial identity could be that the ability of givers and recipients

to view one another in the same room or track each other by name may have engendered

Page 6: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  6  

a sense of social pressure to cooperate or donate more, where a co-ethnic was a potential

borrower. In addition, the reason such identity effects apply to microfinance lending and

trust games could be that contributors may feel that people with whom they share an

identity are lower-risk borrowers, who will generate higher expected returns. As far as we

know, no study has ruled out both mechanisms, nor have studies of ethnic and gender

preference in microfinance had available to them the detailed data on lender

characteristics we collect from our subjects. Our online setting eliminates the social

pressure effect, since the prospective recipients (borrowers) cannot look back at the

subjects (prospective donors). Our experiment also lets us identify whether decrease in

perceived risk is the main factor at work. First, in one condition but not the other, none of

the repayment from borrowers is returned to lenders. By comparing lenders’ choices of

borrowers in these two conditions, we can see whether the lenders show a stronger

preference for certain borrowers when their own money is as risk. Second, our

experiment lets us control for assessments of the likelihood of repayment by borrowers,

as well as perception of their deservingness, both of which we elicit from the

lenders/subjects.

We find evidence of an identity effect: lenders have a bias towards lending to

borrowers of matching ethnicity6 and to borrowers of the same gender. Moreover,

controlling for perceived riskiness and neediness levels does not significantly reduce the

identity effects, nor does the presence of a financial stake A rough calculation suggests

that when lenders have a financial stake (or when lenders have no financial stake), they

are willing to sacrifice 0.36 (0.29) standard deviations of perceived riskiness in order to

                                                                                                               6 As further defined, below.

Page 7: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  7  

lend to a borrower of a matching ethnicity, and they are willing to sacrifice 0.32 (0.55)

standard deviations of perceived riskiness to lend to a borrower of the same gender.

While comparing the choice of borrowers and the loan amount in different

microfinance platforms might be a potentially interesting way to address these issues, we

choose instead to conduct a lab experiment in which subjects make choices at a website

of our own construction. Our website shares some features of a microfinance platform but

has an additional degree of experimenter control and information capture. Our reasons for

conducting an experiment in this way include: (i) it is difficult, if not impossible, to

identify demographic information (including ethnicity and gender), risk-bearing level and

time preference of lenders from the data of existing platforms; (ii) we can structure the

available borrower set in a manner that is infeasible on the existing websites; and (iii) on

the existing websites, researchers have only collected data about those people who visit

and make a loan, but have not collected data about those who may not have heard of or

been motivated to visit the site, or those who visit the sites but do not make a loan. The

motives of inexperienced agents and experienced agents may be very different. For

example, in charity giving, research has found significant differences between cold list

donors and warm list donors (Karlan, List and Shafir, 2011). Our aim is not so much to

learn about the motives of existing lenders to microfinance, but rather to learn how

varying the factors we focus on might affect lending by a broader cross-section of

individuals (including ones who might be drawn to interest-bearing microfinance

products yet to be designed).

Our results are of interest not only in terms of possible ways of attracting more

lenders and funds into microfinance but also insofar as they contribute to the literature on

Page 8: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  8  

social identity. In the former respect, our results suggest that offering interest to lenders

will not only attract more lenders to make a loan, but also boost the average loan amount.

In the latter respect, we provide evidence of positive effect of ethnic/gender ties in a

setting free of mutual visibility, and when controlling for other borrower qualities

including physical attractiveness and weight. These results suggest that riskiness and

neediness, while important, do not explain the tendency for individuals to support

individuals with whom they share certain common features. Thus, regardless of the

financial return condition, matching lenders’ and borrowers’ ethnicities/genders can be an

effective tactic. However, the probability of a borrower receiving a loan given his/her

subjective qualities, such as their perceived neediness level and perceived riskiness level,

might be affected by the repayment condition.

Before concluding this summary, we think it important to mention an important

caveat to bear in mind when interpreting our results regarding crowding out. While our

participants’ philanthropic motivation is never entirely annulled by offering a financial

return, the concern of microfinance website designers that such private gains might turn

off lenders may have greater applicability when lenders perceive their gain to come at the

expense of the poor borrower; our design reduces this perception given that most of the

return can be attributed to the experimenter. We consider the impact of this concern for

our study’s external validity in the concluding section of the paper.

The rest of this paper proceeds as follows: Section 2 describes our experimental

design. Section 3 reports results, and Section 4 is our conclusion.

Page 9: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  9  

2. Methodology

2.1 Experiment procedure

The experiment consists of a set of pre-stage games and a core two-stage game. In

the pre-stage games, we elicit subjects’ time preference and risk preference by two

standard choice list tasks.7 Subjects are informed that with probability 0.1 one of the pre-

stage decisions will be randomly chosen for payoff realization, and that if this occurs,

they will learn the payment result at the end of the experiment, which prevents

contamination of ensuing choices by a wealth effect. The risk aversion task confronts

subjects with 10 rows, in each of which they must choose between an uncertain option

and a certain one. The uncertain option is a lottery consisting of a 90% chance of getting

$30 and a 10% chance of getting $0. The certain option is initially $16, and is increased

by $2 in each subsequent row, reaching a maximum of $34. The great majority of

subjects prefer the uncertain option initially and switch to a certain option at or before the

latter’s maximum. Our measure of risk aversion is directly decreasing in the level of

certain payoff at which a subject switches from the uncertain to certain option. In the time

preference task, subjects are given 9 rows and are asked to choose between two certain

options with different timing. One option is always receiving $10 at the end of the

experiment session. The other option is an amount to be received in 18 months, which is

$10 in the first row and rises by $5 in each subsequent row to a maximum of $50. Our

measure of present time preference is directly decreasing in the level of the first delayed

payoff the subject prefers to receiving $10 the same day.

                                                                                                               7 The instruments used are chosen and revised from Sutter, Kocher, Glätzle-Rützler, and Trautmann (2013). This method has been widely used in experiments, see also Holt and Laury (2002).

Page 10: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  10  

In the first stage of the core experiment, subjects are asked to play a modified

trust game.8 In the game, subjects have a $10 participant fee and an additional $10

endowment that they decide how to allocate. Subjects decide how much of the additional

$10 of endowment money to lend to small business operators, knowing the loan’s

repayment rate and, in some cases, that the borrowers are from poor developing countries

(see section 2.2), but prior to seeing further details about the potential borrowers.

Subjects know that each dollar they decide to lend will be doubly matched by the project,

becoming three loanable dollars. Subjects are also told that the average probability that a

borrower would repay the loan is 90%.9 To avoid having choices of $0 be made with the

sole aim of completing the experiment more quickly, subjects were told (truthfully) they

would need to complete the next stage regardless of their decision.

Having decided how much to lend, subjects learn more about, provide their

impressions of, and rank the prospective borrowers in the second stage of the core

experiment. Specifically, subjects now study photos of 12 possible borrowers who differ

with respect to gender, ethnicity/region (defined here as from Latin America, sub-

Saharan Africa, and Southeast Asia),10 and occupation (farmer or retailer). Each of the 12

                                                                                                               8 We chose to conduct a modified trust game because the microfinance model shares some similar characteristics to a trust game, in which a first-mover (trustor) can send part of his/her endowment to a second-mover (trustee), knowing that the amount he/she sends will be doubly matched by experimenters, and that the second-mover can choose how much to send back. Tripling of returns, as in the original trust or investment game design (Berg, Dickhaut and McCabe, 1995), also makes lending attractive on altruistic grounds. In some modified dictator game experiments, the money giving was matched. See De Oliveira, Croson and Eckel (2011); Eckel, De Oliveira, and Grossman (2007), and so on. 9 We used the 90% figure rather than the 97% figure provided on the website of the intermediary used by us without informing subjects—kiva.org—to eliminate any chance of misleading subjects with an overly optimistic estimate. 10 Latin American borrowers used are from Belize, Costa Rica, El Salvador, Guatemala, Nicaragua, Bolivia, Chile, Colombia, Ecuador, Paraguay, Peru or Mexico; African borrowers are from Burundi, Cameroon, Congo, Ghana, Kenya, Liberia, Mali, Mozambique, Senegal, Sierra Leone, Tanzania, Togo or Uganda; and Southeast Asian borrowers are from Cambodia or Vietnam. We selected the three regions because there were adequate borrowers from each region listed on kiva.org, each group’s members differ identifiably from those of the others with respect to characteristic physical features and names, and each

Page 11: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  11  

types made possible by the crossing of these three dimensions always has one

representative. Subjects see 12 photos at once, and they can click on the photo to read the

narrative of each borrower.11,12 Figure 1 shows an example of how we display the

available borrowers in one of the sessions. After reading each narrative, subjects are

asked to rate (a) the neediness and (b) the riskiness of each of the prospective

borrowers.13 They then choose to which of the borrowers they prefer to lend the loan

amount determined in stage 1, indicating their top three borrowers (in order) out of the 12

choices.14 Once this step has been completed, subjects are asked to provide demographic

information about themselves, yielding the lender characteristics information important to

                                                                                                                                                                                                                                                                                                                                         group also has a corresponding U.S. minority group (as well as some foreign students at the university) with which it has at least superficial linkage—Latin Americans with U.S. Hispanics, Africans with African-Americans, and Southeast Asians with U.S. Asians (especially those of East and Southeast Asian ancestry). We considered the use of East Asian or South Asian borrowers as alternatives but did not find sufficient numbers in kiva.org’s borrower pool. We selected Latin American borrowers who appeared to have Amerindian or mixed race (Mestizo) ancestry in order that there be a sense of ethnic/racial distinction from U.S. students of strictly European ancestry, as with the other two regions. In our analysis of lender-borrower ethnic pairing, we thus have a fourth ethnic/racial category for the U.S., namely White, whose members we thought not likely to perceive themselves as sharing an ethnic/racial identity with any of the three borrower ethnic/racial categories. Self-reported non-Hispanic White subjects chose Latin-American borrowers less often than those of the other two groups, and also less often than did self-reported African-American and African and self-reported Latino lenders, which in conjunction with the ethnic pairing results described below seems consistent with our conjecture. 11 We also partially randomize the order in which borrowers are displayed, according to category, in order to minimize any influence of screen location on choice of borrower. Arrangements are not fully random in that we avoid ones that happens to cluster borrowers of a particular ethnicity or gender in the same part of the screen. 12 As examples of borrowers’ narratives, consider the following: Example 1. Rith is a farmer in Cambodia. She grows maize and rice as the main source of income. She is requesting a loan to buy insecticide for her crop and to pay plowing fees. Example 2. Paul is a business man in Kenya. He sells polythene bags and some other retail commodities. Paul is seeking a loan in order to buy inventory for his shop. 13 The instructions explain that by “neediness” we mean financial need or poverty, not the merit of the borrower’s project as such. To minimize the order effect, subjects decide on each rating for all 12 borrowers before submitting them, and subjects are not allowed to change these answers during the next step, which is selection and ordering of borrowers they wish to lend to. If we allow subjects to change their ranking of neediness and riskiness, they might want to justify their choices. Because of the subjectivity of the each rater’s input, we normalize each subject’s scores to yield a standardized scale across different subjects. For example, let 𝑏!" be the rating for neediness rater i gives to borrower j (j=1,…,N), a normalized

score is generated by 𝑏!!" = !

!"!   !!"/!!!!!!

. 14 To give subjects whose stage 1 lending choice is zero an incentive to make decision in this second stage, we inform them in stage 2 that there is a 1% chance that a $10 loan would be randomly assigned to them and go to the top borrowers they choose.

Page 12: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  12  

our analysis. All subjects, including but not limited to those who might have loan

repayment funds due back to them in 18 months, are also asked to click on a link to a

separate website at which they provide their contact information for receiving a future

payment (if due), with care taken to assure subjects that the personal information

provided could not be linked to their decisions.15

After each session, we ran an algorithm to calculate the exact loan amounts from

lender i to borrower j, based on lenders’ loan amounts and their choice of top three

borrowers. Then, we carried out the loans via kiva.org. We used an algorithm to allocate

the loans because loan amounts at kiva.org need to be at least $25 or in multiples of $25,

and each subject in our experiment could have an integer loan amount from $1 to $30.

(Details are available on request.).16

2.2 Treatment

Our treatments all include the same sets of decisions, but differ in terms of

financial consequences for the lender/subjects and the framing of the core first stage

decision. In the first dimension, we vary the percentage of repayment the subject

                                                                                                               15 An instruction for all treatments can be found in Appendix B1. In the consent form that subjects read before beginning the experiment, they were informed that they would definitely receive a payment at the end of their session and that they might also end up being owed a payment in 18 months (one they later learned could be generated by the time preference task or by making a loan in High- and Medium-Return conditions (see below). To avoid having loan decisions be motivated by the desire to appear pro-social to the experimenters, subjects were told (and end-of-session payout procedure was adapted to add credence to the promise) that their decisions could not be linked to their names and that the personal contact information provided by them for receiving any payment due them in 18 months is handled by an office of the university unrelated to the experiment team. Linkage of loan repayment to specific subjects was achieved by assigning subject ID numbers accessible to both the experiment team and the unrelated administrative staff who kept these personal contact details. 16 Subjects were informed that their loan could not be guaranteed to go to their first, or even their second and third, choice of borrower, but were told all efforts would be made to implement their preference. The detail about the implementation problem being due to a $25 minimum was withheld so as not to diminish effort on the borrower selection task or encourage preference for specific loan amounts. Subjects never visit kiva.org during the experiment and are informed only that the experimenters are working with “an intermediary” to make loans to the individuals depicted in the photos and descriptions shown them.

Page 13: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  13  

receives to examine how financial incentives affect lending decisions. In the High-Return

(HR) condition, 100% of any repayment from the microfinance borrower(s) goes to the

subject. In the Medium-Return (MR) condition, only 50% of any repayment goes to the

subject, with the other half kept by our research project.17 In the Project-Return (PR)

condition, any repayment goes entirely to our project. Note that amounts subjects

allocated for making loans were tripled, so the decision in the modified trust game could

be viewed as an investment decision with a potential 200% interest rate (170% expected

rate, given p = 0.9 repayment probability) for the HR condition, a 50% (35%) interest rate

for the MR condition, and a negative 100% interest rate (or simply a gift with 200%

matching by the experimenter, and the status of loan to the borrower) for the PR

condition. Thus, the lending decision is strictly altruistic in the PR condition, whereas the

decision may be based on a mixture of selfish and altruistic motivations (with potential

influence by time preference and risk bearing preferences) in both the HR and MR

conditions. The range of return rates to the lender—from -100% to 35% or 50% and to

170% or 200%—is probably too broad for making fine policy inferences (a matter to

which we return in the concluding section), but certainly brackets the returns in existing

and conceivable one-to-one microfinance lending.

In the second dimension, we vary the description of the lending opportunity, in

particular how the set of prospective borrowers is described in the initial stage of the core

experiment. In the Neutral Framing (NF) conditions, we tell subjects that the loan they

can make will go to a small business operator in another country. In the Poverty Framing

                                                                                                               17 The instructions are not explicit about what the project will use the funds for. However, since we say “returned to our research project” rather than referencing the broader experimental economics programs at our university, it is possible that participants who have altruistic associations with microfinance may see funds returned to the project as having potential social benefit.

Page 14: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  14  

(PF) condition, we tell subjects that the loan they can make will go to a small business

operator “in poor developing countries such as Bolivia, Kenya and Cambodia,” and we

also give subjects a short description of microfinance, mentioning that the concept’s

“pioneer developer was awarded the Nobel Peace Prize in 2006” and quoting from the

Nobel Committee’s citation (“even the poorest of the poor can work to bring about their

own development”, etc.). We expect that this manipulation of framing increases many

lenders’ intrinsic incentive to make a loan; thus, subjects in PF conditions will loan more

than those in NF conditions, ceteris paribus. The crossing of our three return conditions

and two framing conditions yield the six treatments NF-HR, PF-HR, NF-MR, PF-MR,

NF-PR and PF-PR.

2.3 Structuring 12 borrowers

One borrower of each of the 12 types is chosen by us from among potential

borrowers listed at kiva.org right before each session based on borrowers’ countries,

occupations, and genders.

In the country dimension, we include only Southeast Asian, Sub-Saharan African

and Latin American borrowers. The specific countries represented by a region in a given

session depend on what borrowers kiva.org has available on the given day.

Regarding the borrowers’ occupations, we include only farmers or retailers. With

regards to retailers, we do not include those who sell food or fruit. This is because

subjects may connect the grocery to farmers, and we want to strongly divide agriculture

and retail into different categories. Borrowers in agriculture are mostly from rural areas

and borrowers in retail are mostly from urban areas.

Page 15: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  15  

In addition, we choose borrowers who are between 20 and 50 years old. We

choose borrowers who have sufficient remaining loan need, so that the chance their loan

target will already be met when our session ends (in which case they are no longer

eligible) is kept to a minimum. We crop borrowers’ photos from kiva.org, make the size

of each borrower’s photo uniform (2 inches X 2.4 inches) and show only a borrower’s

shoulders and head for consistency across borrowers (see Figure 1). For each borrower,

we rewrite and formalize his/her narrative to include only occupation, country, and the

purpose of the loan. We delete all other information shown at kiva.org, such as family

structure, loan amount, borrowers’ last name, experience, and income.18

2.4 Third-party evaluation

A possible concern with our experimental design is that choice of borrower might

be influenced by factors other than the three dimensions on which we selected them for

inclusion. To deal with such uncontrolled factors, including the perceived physical

attractiveness of borrowers and whether they are smiling in their photos, we asked five

raters who were not participants in the experiment to review all borrowers’ photographs

and narratives and to quantify certain objective and subjective qualities of each borrower.

Evaluation was conducted between July 29 and August 3, 2014, and between September

16 and September 22, with each rater spending a total of 6 to 9 hours over several days

                                                                                                               18 In addition to the criteria listed, we used discretion when selecting among prospective borrowers by favoring those having clearer photographs displaying features typical of the ethnic and racial grouping in question, e.g. we chose Latin American borrowers striking us as displaying Amerindian or mixed Amerindian and European (Mestizo) features as opposed to features suggestive of strictly European descent. See footnote 10 above.

Page 16: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  16  

(no more than 2 hours per day), to complete the evaluation of 214 borrowers.19 Each rater

was paid a flat fee of $12 per hour.

We included a diverse selection of raters. All raters are undergraduates of Brown

University who vary in ethnicity (Asian or Asian American, African or African

American, Hispanic or Hispanic American, and Caucasian/White) and gender

(male/female).20

Raters are given an instruction from an experimenter prior to their first evaluation.

The instruction for coding is borrowed and revised from Jenq, Pan, and Theseira (2012),

and can be found in Appendix B.2. Raters are told that the qualities they will code

include the borrowers’ attractiveness, happiness, heaviness and so on. Then, raters see all

214 borrowers’ photos. They are told to use a 7-point scale for most of the subjective

qualities, and think of the number 4 as average for each subjective quality. Raters can

review all 214 borrowers’ photos at any point of the evaluation procedure.

3. Experimental Results

3.1 Subject pool and aggregate results

We conducted 25 sessions between May 2013 and September 2014. 295 subjects

participated in the experiments. All the subjects are undergraduate students at Brown

                                                                                                               19 We conducted 25 sessions in total and thus had 25*12=300 borrowers. However, we used some borrowers in more than 1 session, which left us only 214 different borrowers. 20Raters and subjects were both recruited by solicitation emails via the BUSSEL (Brown University Social Science Experimental Laboratory) registration system. Selection from among prospective raters to assure diversity across the categories relevant to the experiment was carried out based on responses to a brief preliminary questionnaire about gender, ethnicity, race and nationality. This questionnaire was sent to responders to the rater solicitation but not to responders to the solicitation of ordinary participants, who received no experiment-specific information before their sessions other than an indication of duration and of minimum and average expected earnings.

Page 17: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  17  

University who still have at least 3 remaining semesters at Brown.21 Each session took

around 75 minutes. We drop the data from 28 subjects because they have inconsistent

choice patterns in one or both of the pre-stage games.22 Thus, we use only 267 subject

observations for the analysis in this paper.

218 out of 267 subjects (81.65%) choose to lend a positive amount rather than

keep all the money available; 73 out of 267 subjects (27.34%) lend all $30. The average

loan amount of these 267 subjects is $14.54, 48.5% of the amount possible. This number

is close to the average amount sent in 84 trust game experiments surveyed by Johnson

and Mislin (2011), exactly 50%.23

We report summary statistics of various characteristics of the subjects by

treatment in Appendix Table A.1. Not surprisingly, given the random assignment, the

subjects in the six treatments look similar along every dimension. Subjects’ demographic

information (except the semester level) is not significantly different between treatments.

                                                                                                               21 When recruiting subjects, we exclude students who have less than 3 remaining semesters at Brown because subjects in our experiment get part of their payment in 18 months, and we believe worry about an office at Brown being able to track him or her down for payment would be smallest among such students. To further check whether the subjects still worry that the payment might not actually made to them, we asked subjects “what % probability you attached to the outcome of not receiving money owed to you” in one of the questions in the post-experiment survey. Result show that the average probability is between 10% and 20% and that the probabilities are not significantly different by treatments (χ2 (5) with ties = 3.468, p-value = 0.6282). 22 28 subjects switch repeatedly between the certain payoff and the uncertain payoff (or payment today and future payments) or choose the uncertain option over “$30 for sure”, which is identical to the lottery’s largest prize. We believe that most of the inconsistencies we observe are due to misunderstanding of the instructions and thus we drop these 28 data in our analysis. 23 One might anticipate higher than typical sending in our own version of the trust game because most subjects are provided with a philanthropic motivation largely lacking in conventional trust games and our “trustees” are explicit loan recipients given various incentives to repay. That our subjects must wait 18 months for repayment and may harbor doubts about our logistical capacity to repay them could be factors biasing lending downwards. Note that average share lent in NF-HR condition, which comes closest to the standard trust game in that it has little philanthropic motivation and full return to the trustor, is 56.3%, while when philanthropic motivation is added, in the PF-HR condition, the share lent rises to 66.1% (for the remaining treatments, see below), well above the literature’s average of 50%, as anticipated.

Page 18: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  18  

We generate the variable “lenders’ ethnicity” by the following procedure. In one

of our post-experiment question, we ask subject whether they are Hispanic, and we also

ask them how they identify themselves. Options (based on U.S. census classifications)

include (1) American Indian or Alaska Native (including all Original Peoples of the

Americas), (2) Black or African American (including Africa and Caribbean), (3) Native

Hawaiian or Other Pacific Islander (Original Peoples), (4) White (including Middle

Eastern), (5) South Asian (India, Pakistan, etc.), (6) East and Southeast Asian (Chinese,

Japanese, Korean, Cambodian, Vietnamese, etc.) and (7) Other.

In the analysis, we divide subjects into five ethnic categories.

(a) Asian (if subjects are not Hispanic and choose (5) or (6) above)

(b) Black (if subjects are not Hispanic and choose (2) above);

(c) Hispanic, Latino or American Indian (if subjects choose (1) above, specify that they

are Hispanic, or choose (7) and specify that they are Mexican or Latino);

(d) Non-Hispanic White (if subjects are not Hispanic and choose (4) above);

(e) Other (if subjects choose none of the above or choose multiple categories).

We list the number of subjects in each ethnic category by gender in Table 1. The

population in our sample is roughly representative of the university’s undergraduate

population with respect to gender and ethnicity.24

                                                                                                               24 In Fall 2013, around 12% of Brown undergraduate students were Hispanic, 7% Blacks/African Americans, 47% non-Hispanic whites, and 18% Asian. Regarding our subject pool, 11% are Hispanic, 9% blacks, 42% are non-Hispanic whites, and 29% are Asian. A binomial test shows that the proportion of Asians in our sample is significantly different from that at Brown at the 1% level, but the proportions of other ethnic categories are not significantly different from the proportions at Brown.

Page 19: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  19  

3.2 Loan amount

In stage 1, each subject can choose an integer loan amount between $0 and $30.

We report summary statistics regarding the percentages of subjects who make a loan of

positive amount and subjects’ loan amounts by treatments in Table 2.

We see that the percentage of subjects who make loans of positive amount

increases with the return rate. We use chi-square tests to compare the fraction of subjects

who make loans of positive amount in different treatments (see Table 3). The results

show that the difference between PF-MR and PF-PR treatments are not statistically

significant, suggesting that being able to earn a return on a loan does not reduce subjects’

incentives to make loans. The results also show that the difference between PF-MR and

PF-HR treatments is not statistically significant, suggesting that being able to earn more

return on a loan does not reduce subjects’ incentives to make loans.

In addition to effects on the percentage of positive loan amounts, the average loan

amount also increases according to the repayment amount, both in the PF condition and

in the NF condition. The average loan amount is $6.81 out of $30 in the NF-PR

treatment, where subjects do not get their principal back. It is $14.30 in the NF-MR

treatment, where subjects receive 1.5 times their principal in 18 months, given that

borrowers repay the loan. A nonparametric Mann-Whitney test based on ranks can be

used to investigate whether the sample of loan amounts comes from populations with the

same distribution. In Table 4, we report the results of a pairwise comparison of the

different treatments. Each number indicates the p-value. The difference between the

distributions in NF-MR and in NF-PR is statistically significantly at the 5% level. The

average is then higher and equal to 16.88 in the NF-HR treatment, where subjects receive

Page 20: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  20  

3 times their principal in 18 months (if repaid). The Mann-Whitney test also shows that

while the difference between the distributions in NF-HR and in NF-MR is not

statistically significant, the difference between the distributions in NF-HR and in NF-PR

is statistically significantly at the 1% level.

That lending also increases with return rate in the PF conditions is clear from

Panel (B). The average loan amount is $11.52 in the PF-PR treatment, $15.06 in the PF-

MR treatment and $19.83 in the PF-HR treatment. The loan amounts are larger in PF-

MR than in PF-PR, although the difference of distribution is not statistically significant,

according to a Mann-Whitney test. The difference between the distributions in PF-HR

and in PF-MR is statistically significantly at the 5% level. The difference between the

distributions in PF-HR and in PF-PR is statistically significantly at the 1% level.

To see whether the results are still robust after we control for subjects’ time

preferences, risk-bearing levels and demographic information, we use OLS regressions

and Tobit regressions. The results shown in Table 5, including Wald tests of difference

between coefficients, are broadly consistent with the results of the Mann-Whitney tests.

Comparing PF and NF conditions holding in each repayment condition shows that

associating the loan opportunity with normatively extolled microfinance clearly boosts

the average loan amount. The average loan amount is higher in PF-HR than in NF-HR,

mostly due to higher loan amounts which are conditional on choosing to lend. The

average loan amount is higher in the PF-MR than in NF-MR, although in this case it is

likely due to a higher percentage of subjects choosing to make a loan. Comparing PF-PR

and NF-PR, the average loan amount is higher in the former than in the latter due to both

factors: an increasing share of subjects who make a loan, and an increasing amount lent

Page 21: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  21  

which is conditional on choosing to lend. According to a Mann-Whitney test, we see that

the difference is significant in the PR condition (i.e., PF-PR sending differs significantly

from NF-PR sending), but is not significant in the two subject repayment conditions (i.e.,

PF-MR sending does not significantly differ from NF-MR sending, nor do PF-HR and

NF-HR sending significantly differ). This suggests that the presence of an intrinsic

incentive plays a stronger role when there is no extrinsic incentive than when substantial

financial returns are present.

Much as the effect of intrinsic incentives is weakened by the presence of

substantial financial returns, so too does the raising of a financial return have less effect

on subject lending when the strong philanthropic motive of a known microfinance

lending opportunity is present. In the NF condition, subjects lend 148% (110%) more

when able to potentially triple (alternatively, add 50% to) their money in 18 months than

when the loan is a donation without recouping of principal. Subjects in the PF condition,

in contrast, lend only 72% (31%) more in the High-Return (Medium-Return) than in the

Project-Return condition. In Table 5, the coefficient of PF-MR indicates that the change

from the NF-PR to the NF-MR treatment is larger than that from PF-PR to PF-MR, and

the difference is statistically significant at the 5% level only after we control for risk-

bearing level, time preference level and other demographic information. The coefficient

of PF-HR indicates that the NF-PR to NF-HR changes more than it does from PF-PR to

PF-HR, although the difference is not statistically significant.

In addition, the OLS regression results in Table 5 allow us to compare the effect

of a pure intrinsic incentive with the effect of a pure financial inducement. Subjects who

are given a 50% (medium) return on their loan as extrinsic incentive but who have a

Page 22: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  22  

weaker philanthropic motivation lend almost twice as much as do subjects whose loans

are strictly a gift but who are provided with a stronger philanthropic motivation (that is,

NF-MR vs. PF-PR -- 8.60 vs. 4.32, significantly different at the 10% by Wald test),

controlling for subjects’ demographic information, risk-bearing level and time-preference

level. Subjects who are given a 200% (high) return on their loan as extrinsic incentive but

have a weaker philanthropic motivation lend more than twice as much as do subjects

whose loans are strictly gifts but who are given a stronger philanthropic motivation (that

is, NF-HR vs. PF-PR -- 10.81 vs. 4.32, significantly different at the 1% level by Wald

test), controlling for subjects’ demographic information, risk-bearing level, and time-

preference level.

The results also show that measured time preference is a significant determinant

of loan amount, but measured risk-aversion is not. Consider two subjects: one subject

asked for $1 more interest for a $10 loan (i.e., asked for 10% more interest rate) than the

other subject. For convenience, we call the former one an “impatient subject”, and we

call the latter one a “patient subject”. Suppose that both subjects are considering how

much to lend for 18 months. The result in Equation (3) in Table 5 shows that holding the

rate of return constant, the “patient subject” loans $0.39 more than the “impatient

subject”, the difference being statistically significant at the 1% level.25

3.3 Subjects’ selection of their top three borrowers

In stage 2, subjects can choose their top 3 preferred borrowers to whom they

would like their chosen loan amount to go. Note that we structured the 12 borrowers by

ethnicity, gender and occupation in each session, thus in the chi-square test we treat each

                                                                                                               25 Note that this result comes from the pooled data (i.e., including all treatments). If we divide the data by the repayment condition, we can see that when providing a -100% (50%) [200%] interest rate, the patient subject loans $0.14 ($0.45) [$0.44] more than the impatient subject.

Page 23: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  23  

borrower with the same ethnicity, gender and occupation as the same type. Note also that

at this stage the difference between Poverty Framing condition and Neutral Framing

condition no longer matters because all subjects see the same information about the

potential borrowers and know that they are small business operators in developing

countries. In the analysis, we pooled the data in these two conditions. Additionally, we

pooled the data of the Medium Return condition and the High Return condition because

in both conditions subjects have the possibility to receive repayment.26 In sum, in the

second-stage analysis, our treatment alternatives are Subject-Return (if HR and MR in

both framing conditions) and Project-Return (if PR in both framing conditions).

The main question we are interested in here is whether lenders prefer borrowers

of the same ethnicity/gender as themselves more than lenders of a different

ethnicity/gender do, and whether the strength of this preference (if present) depends on

the repayment condition. Note that in some cases—for example, the correspondence

between self-identified African-American lenders and sub-Saharan African borrowers—

the term “matching ethnicity” is more accurate than “same ethnicity,” and that we use the

latter for convenience, relegating further concerns to a note.27

                                                                                                               26 We carried out preliminary analysis and found that subjects in HR condition did not choose significantly different borrowers than subjects in MR condition. 27 We face numerous such problems of imperfect correspondence. In order to have sufficient number of lenders (U.S. university student subjects) and borrowers in small numbers of common ethnic categories, we must treat African-Americans whose most recent African-born ancestor arrived in the U.S. more than a century-and-a-half ago as members of the same group as the children of recent African immigrants to the U.S. and as African students studying abroad in the U.S., treating all jointly as being of the “same” ethnicity or race as the experiment’s sub-Saharan African borrowers. Parallel issues exist for the Asian and Latino or Hispanic categories. The problem of heterogeneity within putatively homogeneous categories may be exacerbated if individuals from different sub-regions, for example South Asians and Southeast Asians, do not perceive themselves as sharing an ethnicity, whereas our approach treats U.S. students of all Asian (likewise all sub-Saharan, all Latin American) backgrounds as belonging to a common category. In one of the few cases in which our sample contains sufficient data for formal testing of within-ethnicity differences, a chi-square test shows that the percentage of South Asians favoring Southeast Asians is not significantly differently from the percentage of East and Southeast Asians favoring Southeast Asians (χ2(1)= 0.0779).

Page 24: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  24  

In table 6, we report the numbers (percentage) of subjects who prefer a specific

ethnicity of borrowers based on their 1st choice of borrower. The columns indicate the

ethnicity of the borrowers and the rows that of the lenders. We find that lenders overall

show some preference among the three ethnicities,28 but since our numbers of Asian,

Black, Latino and non-Hispanic White lenders vary considerably, it is more illuminating

to look at lender-borrower pairing effects. We find that lenders prefer borrowers of their

own ethnicity more than lenders of other ethnicities do. Whereas slightly more than 1/3

of lenders prefer Asian borrowers when all lenders are pooled, around ½ of the Asian

lenders prefer Asian borrowers, which is significantly more than for non-Asian lenders at

the 1% level ((χ2(1)= 7.05). Around ½ of African-American and African lenders prefer to

lend to African borrowers, which is more than the average lenders do. We can also see

that around 1/3 Hispanic lenders prefer Hispanic borrowers. Although the difference in

the Black and African case is not statistically significant and that in the Hispanic case is

not significant when all treatments are pooled, the latter difference is significant at the

10% level (χ2(1)= 2.96) in the Project-Return treatment.

Another potentially important factor that might influence the choice of borrower

is the gender of the lender vs. the gender of the borrowers. In Table 7, we report the

number (percentage) of subjects who most prefer lending to a borrower of a certain

gender based on the 1st choice of borrower. The columns indicate the genders of the

borrowers and the rows the genders of the lenders. We find that both female and male

lenders prefer female borrowers, especially in the Project-Return treatment. While male

                                                                                                               28 Taken together, 39% of subjects prefer Asian borrowers, 34% prefer African borrowers and 27% prefer Hispanic borrowers. Binominal test rejects the hypotheses that the probability of most preferring Asian borrowers and that the probability of most preferring Hispanic borrowers equal 1/3 in all treatments combined at the 5% level and in the Project-Return treatments (pooled NF-PR and PF-PR) at the 10% level for Asian and at the 1% level for Hispanic.

Page 25: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  25  

lenders slightly prefer female borrowers to males, selecting a female borrower in 52% of

their first choices, female lenders prefer female borrowers by a significantly larger

margin: 74% of first choices. This difference is significant at the 1% level (χ2(1) = 14.14)

for all treatments when pooled, at the 1% level (χ2(1) = 9.34) for the pooled Subject-

Return treatments, and at the 5% level (χ2(1) = 4.12) for the Project-Return treatments.

While the results by chi-square test are potentially interesting, each borrower’s

specific characteristics are not taken into account in this non-parametric test. Might there

be other variables that are correlated with ethnicity and gender, driving the result?

To control for other factors that might affect borrower choice, we use a rank-

ordered logit model to estimate how subjects value borrowers’ characteristics. We gather

each borrower’s objective/subjective qualities data by taking the average of our 8 raters’

evaluations, and we also use the subjects’ responses in the experiment regarding the

perceived neediness level and perceived riskiness level of each borrower. We generate a

standardized perceived neediness/riskiness level for each borrower and lender by

subtracting the raw value from a subject’s average perceived level across all N borrowers

(N=12), and then dividing this value by the standard deviation (𝑏!!" = !

!"!   !!"/!!!!!!

).

Since we have limited sample size for some ethnicities (e.g. in the project-return

treatments we have fewer than 10 observations for African/African-American and only

around 10 observations for Hispanic subjects, we use dummy variables, "lenders and

borrowers are of the same ethnicity" and "lenders and borrowers are of the same gender"

instead of all the interactions. Our model can be written as

value (borrower j by lender i) = βXj+ γGij + λRij + εi

Page 26: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  26  

where value is the preference rank i gives j, β is a 1x N vector which captures the

coefficient of Xj, a Nx1 vector that captures all borrower characteristics (including

neediness and riskiness as rated by i and attractiveness and other qualities as judged by

the raters), γ is a scalar which capture the coefficient of Gij, a dummy variable which

captures whether lender and borrower are of the same gender, λ is a scalar which captures

the coefficient of Rij,a dummy variable which captures whether lender and borrower are

of the same ethnicity, and εi is a random residual capturing any omitted characteristics.

We assume the εi are independent and follow an extreme value type I distribution.

Note that in our experimental design, each lender sees 12 structured borrowers

and chooses and orders his/her 3 most preferred borrowers. Thus, each lender generates

12 observations, with the dependent variable set to 3 for most-preferred, 2 for second-

most-preferred, 1 for third-most-preferred, and 0 for those failing to receive a top three

ranking. In a rank-ordered logit model, each borrower’s value is assumed to be linear in

the characteristics, with the coefficients expressing the direction and weight of the

characteristics. Assuming that lender i is comparing all 12 borrowers (j=1,…,12), the

probability that borrower 1 is ranked ahead of all the other borrowers can be written in

the multinomial logit form:

Π1 = Pr {value1} > max (value2, value3, …., value12)} = !!"#$%!

!!"#$%&!"!!!

The results are shown in Table 8. In specification (1), we can see that subjects

prefer borrowers of the same ethnicity and the same gender. However, this positive effect

might be caused by “trust”. That is to say, subjects may perceive borrowers of the same

ethnicity/gender as more trustworthy than the other borrowers and may simply prefer a

borrower with lower perceived default probability. To explore this concern, we add a

Page 27: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  27  

control for borrowers’ perceived riskiness levels in specification (2). Not surprisingly, we

find that subjects prefer to lend to borrowers they judge to have a higher likelihood of

repayment. However, we find that the coefficients on shared ethnicity and shared gender

are little changed in value or significance after we add this control, implying that the

favoring of same ethnicity and same gender borrowers is not driven by a “trust my

brothers” effect.

A glance at Table 8 shows that significant coefficients on other variables in

regressions (1) and (2) vary in their robustness to the addition of further controls. In

specification (3), we add controls for j’s neediness level (as perceived by lender i) and

average perceived attractiveness level, perceived happiness level and weight (judged by

the raters). Two results not yet mentioned appear in the initial specifications and prove

robust to the new controls: that lenders pooled across treatments assign higher value to

female borrowers than to male borrowers, and to farmers than to business women/men.

Two other coefficients are initially significant but not robust: general biases towards

Asian and African borrowers, irrespective of own ethnicity, appear present in

specifications (1) and (2) but become insignificant (and in the case of Asian borrowers

flip sign) when the new controls are added. In addition, lenders prefer to lend to

borrowers they perceived as more needy and less risky, and borrowers who appear (to the

evaluators) to be less heavy (more slim).29 These results are in line with the empirical

evidence in Jenq, Pan and Theseira (2013).30 We can also see that lenders prefer

                                                                                                               29 We captured the display location by 11 separate dummies for all possible positions except an omitted category. The result is similar if we add these 11 separate dummies for the borrowers’ display locations, which does not in itself show significant explanatory power. 30 Jenq, Pan and Theseira (2013) found evidence at kiva.org that the borrowers who appear (to the evaluators) to be more attractive, less obese, more needy, more honest and more creditworthy were funded more quickly.

Page 28: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  28  

borrowers of the same ethnicity and/or of the same gender, with other things equal. Being

the same ethnicity or being the same gender are as important to lender choice as is a 0.40

standard deviation decline in the perceived riskiness level or a 0.32 standard deviation

increase in perceived neediness level.31

In specification (4), we consider a final amendment of our rank-ordered logit

model to explore whether subjects in all repayment conditions use the same valuation

function. That is to say, we are interested in whether lenders alter their choice of

borrowers when all the repayment from the borrowers are returned to the experimenters,

not the lenders. To investigate this, we add interaction terms between all control variables

and a “ProjectReturn” dummy variable (PR). If the coefficients are significantly different

from 0, we know that these attributes affect borrower choice differently depending on the

repayment condition. Specifically, we consider a model as follows:

value (borrower j by lender i) = βXj+ γGij + λRij

+ α *PRj * [βXj+ γGij + Rij ]+εi

where α is a 1x (N+2) vector which captures the coefficient of PRj, an (N+2)x1

dummy vector.

Turning to the results, we see that the interactions of PR with same gender and

same ethnicity are statistically insignificant, suggesting that these effects hold regardless

of return condition, although their levels of statistical significance are now lower. In

addition, the positive effect of perceived neediness is significantly larger and the effect of                                                                                                                31 A 1 standard deviation increase in perceived neediness level will increase the borrowers’ value by 0.78 (=0.809*0.96), being the same ethnicity will increase the borrower’s value by 0.25. Thus, being the same ethnicity is as important as being 0.32 (=0.25/0.78) standard deviation higher in the perceived neediness level. A 1 standard deviation decrease in perceived riskiness level will increase the borrowers’ value by 0.62 (=0.65*0.96), being the same ethnicity will increase the borrower’s value by 0.25. Thus, being the same ethnicity is as important as being 0.40 (=0.25/0.62) standard deviation lower in the perceived riskiness level. Being the same gender has the same calculation and result as being the same ethnicity.

Page 29: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  29  

physical attractiveness becomes significantly negative, in the PR condition, while the

effect of borrower weight becomes insignificant in the new specification.

The results shown in Table 8 allow us to estimate the probability of lender i

preferring a specific borrower j. Consider a case in which a lender chooses between a

female borrower and a male borrower. These two borrowers are identical in all

subjective/objective qualities (all control variables listed in column (3) of Table 8) and

only differ in gender. Suppose the lender is a female. Then, the probability that the

female lender prefers the female borrower rather than the male borrower is 66%. Suppose

the lender is a male. Then, the probability that the male lender prefers the female

borrower rather than the male borrower is 53% (estimated by the coefficient in column

(3) of Table 8).32 Thus, a female lender prefers to lend to a female borrower 25% more

than does a male lender, and a male lender prefers to lend to a male borrower 38% more

than does a female lender. These magnitudes are not significantly affected by the

repayment condition.

Consider a case in which a lender chooses between three borrowers who are

identical in all subjective and objective qualities (all control variables listed in column 3

of Table 8) and only differ in ethnicity (Asian, African or Hispanic). Suppose the lender

belongs to one of the corresponding ethnic groups (Asian/Asian-American,

African/African-American or Latin American/Hispanic), the probability that this lender

prefer the borrower of his/her ethnicity rather than borrowers of different ethnicity is 18%

greater than a white lender’s preference for a borrower of that ethnicity, and 30% greater

                                                                                                               32 We denote the probability of lender i preferring a specific borrower j by 𝑝𝑟𝑜𝑏!

!, i =female or male and j = female or male. 𝑝𝑟𝑜𝑏!!!"#$%"

!!!"#$%"= exp^(0.374+0.273)/[exp^(0.374+0.273)+1] = 1.91/2.91=0.66; 𝑝𝑟𝑜𝑏!!!"  

!!!!"#$! = Exp^(0.374)/[exp^(0.374)+ exp^(0.273)] = 1.45/2.78=0.53.

Page 30: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  30  

than the preference for that ethnicity by non-white lenders of either of the remaining two

ethnicities (estimated by the coefficient in column (3) of Table 8).33 The magnitude

differs slightly and insignificantly by the repayment condition.

4. Conclusion

Microfinance refers to financial services meant to help low-income individuals

become self-supporting entrepreneurs. Most studies focus on the effects of microfinance

on borrowers (the demanders’ side), while a few studies investigate the factors that affect

lending decisions of existing lenders (the suppliers’ side). One of the factors these studies

investigate is criteria existing lenders use to determine whether or not to lend money to a

borrower. We linked the microfinance model with a modified trust game experiment to

explore the roles of financial inducement and intrinsic incentives. Our study combines

insights gained across the economics literature about the suppliers’ side of microfinance,

as well as the suppliers’ side of philanthropic donation. By introducing a large private

incentive to lend in the Subject Return (i.e. High-Repayment and Medium-Repayment

treatment) conditions, we explored the potential of peer-to-peer lending to attract more

funds if a financial return goes to the lender. By providing different levels of description

of microfinance borrowers in different repayment conditions, we explored whether the

positive effect of the intrinsic incentives would be diminished or even eliminated by the

presence and the strength of extrinsic incentive. By carefully structuring the list of

borrowers, we explored the role social connection plays. Our novel experimental design

                                                                                                               33 We denote the probability of lender i preferring a specific borrower j by 𝑝𝑟𝑜𝑏!

!, i = Asian, African, Hispanic, white, or the others. j = Asian, African or hispanic. 𝑝𝑟𝑜𝑏!!!"#!$

!!!"#!$= 𝑝𝑟𝑜𝑏!!!"#$%!&!!!"#$%!&=

𝑝𝑟𝑜𝑏!!!!"#$%!&!!!!"#$%!&= exp^(0.247)/[exp^(0.247)+2] = 0.39. 𝑝𝑟𝑜𝑏!!!"#!$

!!!"#$%!& = 𝑝𝑟𝑜𝑏!!!"#!$!!!!"#   = 𝑝𝑟𝑜𝑏!!!"#$%!&

!!!"#!$ = 𝑝𝑟𝑜𝑏!!!"#$%!&

!!!!"#$%!&= 𝑝𝑟𝑜𝑏!!!!"#$%!&!!!"#!$ =  𝑝𝑟𝑜𝑏!!!!"#$%!&

!!!"#$%!&=1/[exp^(0.247)+2] = 0.30. 𝑝𝑟𝑜𝑏!!!!!"#!!!"#!" =

𝑝𝑟𝑜𝑏!!!!!"#!!!"#$%! = 𝑝𝑟𝑜𝑏!!!!!"#

!!!!"#$%!& =  𝑝𝑟𝑜𝑏!!!"!!"#!!!"#!$ = 𝑝𝑟𝑜𝑏!!!"!!"#

!!!"#$%!& = 𝑝𝑟𝑜𝑏!!!"!!"#!!!!"#$%!&=1/3.

Page 31: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  31  

also allows us to investigate these effects of ethnic and gender ties in a more remote

setting than in previous laboratory experiments. In addition, it enables us to explore

whether the factors affecting lenders’ choice of borrowers operate differently depending

on whether the possibility of receiving repayment is present.

The data from our experiment permitted us to explore a number of the preference

trade-offs affecting subjects’ lending decisions. For example, we find that at least

implicitly subjects view ability to help a needier borrower as compensation for additional

risk of default, and demonstrate that willingness to trade value of help (“neediness”) for

lower default probability (“riskiness”) is only about half as great when the lender stands

to personally capture any earned return. We demonstrate preference for borrowers of

same gender and same or at least corresponding ethnicity even when there is no direct

connection (mutual visibility, same room presence) between lender and borrower, and

provide estimates of willingness to trade these perceived goods of common identity for

reduced risk, aid to needier individuals, and other valued attributes. Preference for own

ethnicity and gender is not so strong as to overwhelm all other dimensions of the choice,

but remains statistically significant and robust to consideration of numerous factors. We

may be the first to show formally that this preference in the case of a one-to-one lending

relationship is not explained by associated beliefs about repayment prospects.

With regard to practical lessons for microfinance, we regard our experiment as a

small step toward the larger goal of understanding whether and how some of the unique

features of microfinance lending can be harnessed to scale up the flow of credit from

global rich to global poor. These features we explore include (a) the latent benevolence of

rich country savers, (b) the extremely low financial returns on conventional savings

Page 32: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  32  

vehicles, (c) the ability of small developing country entrepreneurs to repay loans at the

rates of interest standard in microfinance, which are high in comparison with returns (b)

but low compared to money-lender and other informal market rates. Our results suggest

that many rich country residents who have not sought out microfinance lending

opportunities may have a substantial latent willingness to lend to small entrepreneurs in

“the global South,” that they are willing to bear some risk to do this, and that the

possibility of financial returns can be complementary to, rather than crowd out, the

appeal of philanthropic motivation. We need to be quite careful in drawing operational

conclusions, in part because we studied only unrealistically high return rates. In addition,

because the lion’s share of the financial return on lenders’ loans can be attributed to the

experimenter, our set-up minimizes any guilt lenders may feel about profiting from the

poor borrowers’ efforts.34 This does not render our results entirely irrelevant, in our view,

because we suspect that sharing with lenders a small fraction of returns now captured by

intermediaries does not need to cause paralyzing guilt, especially in current savers who

are not the “hard core” constituency of philanthropically focused microfinance

platforms.35 Clearly, much additional research is required to investigate these issues.

                                                                                                               34 Other reasons for caution include the usual concerns about laboratory experiments with elite university student subjects, such as the fact that these students are relatively well-informed and privileged and that the intensely scrutinized setting of a laboratory decision-making experiment may favor decisions different from those in ordinary life (List, 2007). 35 Consider, for example, the possibility that a large bank or company such as Walmart, Amazon.com, Paypal or Google, decided to make available to its customers a non-risk-bearing savings vehicle paying the saver, say, 0.8% interest, truthfully advertising that the funds are lent through intermediaries to microfinance borrowers. Suppose that enough mid- to large-sized microfinance organizations in the developing world would work with this company and share with it 1% (out of, say, 30%) of borrower repayments, when these are made. We agree that philanthropically-oriented individuals currently attracted to platforms like kiva.org would not prefer the (say) Amazon.com alternative, with the 0.8% interest perhaps even serving as a discouragement to them, but a large remaining group of customers might see both the interest premium over their conventional checking account and the fact that the accounts help put credit in the hands of credit-constrained micro-entrepreneurs as attractions.

Page 33: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  33  

Our novel experimental design provides us a better understanding of the

possibilities for expanding microfinance. In addition, there are some other analyses we

plan to do, including (1) how time preference and risk-bearing level could affect peoples’

lending decisions and choice of borrowers, (2) how borrowers’ characteristics affect their

chance of being preferred by lenders and (3) whether the preference of similar ethnicities

will be changed by lenders’ place of birth. More experimental and empirical studies are

expected in the near future.

Page 34: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  34  

Reference

Ariely, Dan, Anat Bracha, and Stephan Meier. "Doing good or doing well? Image motivation and monetary incentives in behaving prosocially." The American Economic Review (2009): 544-555.

Berg, Joyce, John Dickhaut, and Kevin McCabe, 1995, “Trust, Reciprocity and Social History,” Games and Economic Behavior 10: 122-42.

Chen, Roy, et al. "Does Team Competition Increase Pro-Social Lending? Evidence from Online Microfinance." (2014).

Choo, Jaegul, et al. "Understanding and promoting micro-finance activities in kiva. org." Proceedings of the 7th ACM international conference on Web search and data mining. ACM, 2014.

Deci, Edward L. "Effects of externally mediated rewards on intrinsic motivation."Journal of personality and Social Psychology 18.1 (1971): 105.

De Oliveira, Angela, Rachel TA Croson, and Catherine Eckel. "The giving type: Identifying donors." Journal of Public Economics 95.5 (2011): 428-435.

Eckel, Catherine C., and Philip J. Grossman. "Altruism in anonymous dictator games." Games and economic behavior 16.2 (1996): 181-191.

Eckel, Catherine C., Angela De Oliveira, and Philip J. Grossman. "Is more information always better? An experimental study of charitable giving and Hurricane Katrina." Southern Economic Journal 74.2 (2007).

Fershtman, Chaim, and Uri Gneezy. "Discrimination in a segmented society: An experimental approach." Quarterly Journal of Economics (2001): 351-377.

Fong, Christina M., and Erzo FP Luttmer. "What determines giving to Hurricane Katrina victims? Experimental evidence on racial group loyalty." American Economic Journal: Applied Economics 1.2 (2009): 64.

Fong, Christina, and Felix Oberholzer-Gee. "Truth in giving: Experimental evidence on the welfare effects of informed giving to the poor." Journal of Public Economics 95.5 (2011): 436-444.

Frey, Bruno S. "How intrinsic motivation is crowded out and in." Rationality and Society 6.3 (1994): 334-352.

Frey, Bruno S., and Felix Oberholzer-Gee. "The cost of price incentives: An empirical analysis of motivation crowding-out." The American economic review (1997): 746-755.

Page 35: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  35  

Galak, Jeff, Deborah Small, and Andrew T. Stephen. "Microfinance decision making: A field study of prosocial lending." Journal of Marketing Research48.SPL (2011): S130-S137.

Glaeser, Edward L., et al. "Measuring trust." Quarterly Journal of Economics (2000): 811-846.

Gneezy, Uri, and Aldo Rustichini. "Pay enough or don't pay at all." Quarterly journal of economics (2000): 791-810.

Gneezy, Uri, Stephan Meier, and Pedro Rey-Biel. "When and why incentives (don't) work to modify behavior." The Journal of Economic Perspectives (2011): 191-209.

Holt, Charles A., and Susan K. Laury. "Risk aversion and incentive effects."American economic review 92.5 (2002): 1644-1655.

Jenq, Christina, Jessica Pan, and Walter Theseira. "What Do Donors Discriminate On? Evidence from Kiva. org." (2012).

Johnson, Noel and Alexandra Mislin, 2011, “Trust Games: A Meta-analysis,” Journal of Economic Psychology 32: 865 – 889.

Karlan, Dean, John A. List, and Eldar Shafir. "Small matches and charitable giving: Evidence from a natural field experiment." Journal of Public Economics95.5 (2011): 344-350.

Lane, Robert E. The market experience. Cambridge University Press, 1991.

List, John, 2007, “Field Experiments: A Bridge between Lab and Naturally Ooccurring Data,” B.E.Journal of Economic Analysis and Policy 5 (2): 1 – 47.

Liu, Yang, et al. "I loan because...: Understanding motivations for pro-social lending." Proceedings of the fifth ACM international conference on Web search and data mining. ACM, 2012.

Morduch, Jonathan, 2013, “How Microfinance Really Works,” Milken Institute Review, 2nd Quarter, 51 – 59.

Rotemberg, Julio J. "Charitable giving when altruism and similarity are linked." Journal of Public Economics (2012).

Ryan, Richard M., and E. L. Deci. "Intrinsic motivation and self determination in human behavior." New York (1985).

Sutter, Matthias, et al. "Impatience and uncertainty: Experimental decisions predict adolescents' field behavior." The American Economic Review 103.1 (2013): 510-531.

Page 36: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  36  

Titmuss, Richard. "The gift of blood." Society 8.3 (1971): 18-26.

Page 37: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  37  

Figures and Tables

Figure 1. An example of how we display the 12 borrowers in the experiment36

                                                                                                               36  In order to keep these borrowers anonymous, we replaced their names displaying here with common names in their countries. However, we did not change their name from names at kiva.org in the on-site experiment.

Page 38: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  38  

Table 1

Number (percentage) of observation by ethnicity and by gender

Ethnicity \ Gender Male Female Total

(a) non-Hispanic Asian 40 38 78 (29%)

(b) non-Hispanic Black 12 13 25 (9%)

(c) Hispanic/ Amerindian 17 13 30 (11%)

(d) non-Hispanic White 61 53 114 (43%)

(e) Others 8 12 20 (7%)

Total 138 129 267

Page 39: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  39  

Table 2

(A) Summary Statistics for the loan amount in Neutral Framing conditions, by repayment

condition

Project Return

(NF-PR)

Medium Return

(NF-MR)

High Return

(NF-HR)

Average 6.81 14.30 16.88

Standard deviation 9.85 12.86 9.88

Median 3 12 15

Positive Loan 55.56% 74.07% 92.00%

Average loan,

conditional on

positive loan

12.27 19.3 18.35

Observation 27 27 25

(B) Summary Statistics for the loan amount in Poverty Framing conditions, by repayment

condition

Project Return

(PF-PR)

Medium Return

(PF-MR)

High Return

(PF-HR)

Average 11.52 15.06 19.83

Standard deviation 10.35 11.78 10.50

Median 9 15 21

Positive Loan 79.37% 84.85% 91.53%

Average Loan,

conditional on

positive loan

14.52 17.75 21.67

Observation 63 66 59

Page 40: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  40  

Table 3

Chi-square test with Pairwise Comparisons of percentage of subjects who

made a loan of positive amount By Treatment

(A) Comparison of percentage of subjects who made a loan of positive

amount in different repayment condition in Neutral Framing conditions

NF-PR NF-MR

NF-MR .154 ---

NF-HR .003*** .088*

B) Comparison of percentage of subjects who made a loan of positive amount

in different repayment condition in Poverty Framing conditions

PF-PR PF-MR

PF-MR .416 ---

PF-HR .058* .251

C) Comparison of percentage of subjects who made a loan of positive amount

in the same repayment condition and in different framing condition

Project Return (PR) Medium Return (MR) High Return (HR)

.021** .222 .943

Page 41: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  41  

Table 4

Mann-Whitney U Tests Based on Ranks with Pairwise Comparisons of

Medians of Loan Amount by Treatment

(A) Comparison of Loan amount between subjects in different repayment

condition in Neutral Framing conditions

NF-PR NF-MR

NF-MR .0326** ---

NF-HR .0006*** .3470

(B) Comparison of Loan amount between subjects in different repayment

condition in Poverty Framing conditions

PF-PR PF-MR

PF-MR .1056 ---

PF-HR .0001*** .0229**

(C) Comparison of Loan amount between subjects in the same repayment

condition and in different framing condition

Project Return (PR) Medium Return (MR) High Return (HR)

.0181** .6627 .1987

Page 42: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  42  

Table 5 Determinants of loan amount

(1) (2) (3) Poverty Framing (PF) 4.71** 5.31** 4.36* (2.29) (2.51) (2.51)

High Return (HR) 10.07*** 10.96*** 10.89*** (2.72) (2.78) (2.66)

Medium Return (MR) 7.48** 7.89*** 8.90*** (3.09) (3.11) (2.78)

PF-HR -1.76 -3.32 -3.44 (3.31) (3.38) (3.23)

PF-MR -3.95 -5.65 -7.28** (3.66) (3.87) (3.52)

Future equivalents for $10 today (from 12.5 to 52.5)

-0.39*** (0.06)

Cash equivalents for a 90%*$30 lottery (from 8 to 29)

0.05 (0.16)

Constant 6.82*** 9.37*** 18.05*** (1.88) (2.81) (6.33)

Demographic control no yes yes Observations 267 267 267 R-squared 0.1134 0.1552 0.2936

p-value for the wald test H0: HR+PF-HR=MR+PF-MR 0.0178 0.0087 0.0010 H0: MR+PFMR=0 0.0720 0.2898 0.4219 H0: HR+PFHR=0 0.0000 0.0001 0.0000 H0: MR=HR 0.4116 0.3058 0.4658 H0: PF+PFHR=0 0.2182 0.4062 0.6749 H0: PF+PFMR=0 0.7892 0.9016 0.2277 H0: PF =MR 0.3202 0.3306 0.0629 H0: PF=HR 0.0238 0.0185 0.0060 Note: OLS regressions. A tobit regression yields similar results. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Demographic information includes lenders’ gender, ethnicity, semester level, whether lender has recently made a charitable donation and frequency of doing volunteer work. None of these demographic controls are significant, except that in equation (2) Hispanic lenders lend $5.19 less, significant at the 1% level, and subjects lend $0.97 less for each additional semester completed, significant at the 10% level. The control variables for time-preference and risk-bearing are defined as follows: Future equivalents for $10 today: the payment in 18 months that the

Page 43: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  43  

subjects values equally to receiving $10 today. We calculate the subject’s future equivalents by taking the midpoint between the two future payments where subjects switched from the money today to the money in 18 months in the pre-stage game. This method was used in Sutter et al. (2013). For subjects who always choose the payment today in the pre-stage game, we code the future equivalents as 52.5. The more the future equivalents, the more impatient subjects are. Very impatient: this is a dummy variable for subjects who always choose the payment today in the pre-stage game. Cash equivalents for a 90%*$30 lottery: a certain payment that subject value equally as a lottery with 90% chance of receiving $30. We calculated the subject’s certainty equivalents by taking the midpoint between the two certain payments where subjects switched from the uncertain option to the certain option in the pre-stage game. This method was used in Sutter et al. (2013). For subjects who always choose a certain option in the pre-stage game, we coded the certainty equivalents as 8. The more the cash equivalents, the more risk loving subjects are. 69.29% of our subjects are risk averse, 20.97% of our subjects are risk neutral and 9.74% of our subjects are risk loving.

Page 44: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  44  

Table 6

Lenders’ top 1 choice, by lenders’ and borrowers’ ethnicity

(A) All treatment

Lenders’

ethnicity

Borrowers’ ethnicity

Asian

borrowers

African

borrowers

Latino

borrowers

Total

(a) Asian 40 (51%) 22 (28%) 16 (21%) 78

(b) Black 5 (20%) 11 (44%) 9 (36%) 25

(c) Latino 11 (37%) 9 (30%) 10 (33%) 30

(d) White 42 (46%) 40 (35%) 32 (28%) 114

(e) Others 6 (30%) 9 (45%) 5 (25%) 20

Total 104 (39%) 91 (34%) 72 (27%) 267

(B) Subject return treatment

Lenders’

ethnicity

Borrowers’ ethnicity

Asian

borrowers

African

borrowers

Latino

borrowers

total

(a) Asian 21 (46%) 12 (26%) 13 (28%) 46

(b) Black 3 (17%) 8 (44%) 7 (39%) 18

(c) Latino 7 (37%) 6 (32%) 6 (32%) 19

(d) White 30 (38%) 22 (28%) 27 (34%) 79

(e) Others 5 (33%) 7 (47%) 3 (20%) 15

Total 66 (37%) 55 (31%) 56 (32%) 177

Page 45: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  45  

(C) Project Return treatment

Lenders’

ethnicity

Borrowers’ ethnicity

Asian

borrowers

African

borrowers

Latino

borrowers

total

(a) Asian 19 (59%) 10 (31%) 3 (9%) 32

(b) Black 2 (29%) 3 (43%) 2 (29%) 7

(c) Latino 4 (44%) 3 (33%) 4 (36%) 11

(d) White 12 (34%) 18 (51%) 5 (14%) 35

(e) Others 1 (14%) 2 (29%) 2 (40%) 5

Total 38 (42%) 36 (40%) 16 (18%) 90

Page 46: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  46  

Table 7

Lenders’ top 1 choice borrower, by lenders’ and borrowers’ gender

Panel A. All treatments

male borrowers female borrowers all

male lenders 66 (48%) 72 (52%) 138

female lenders 33 (26%) 96 (74%) 129

all lenders 99 (37%) 168 (63%) 267

Panel B. Subject Return treatments (MR and HR treatments)

male borrowers female borrowers

male lenders 49 (51%) 47 (49%) 96

female lenders 23 (28%) 58 (72%) 81

all lenders 72 (41%) 105 (59%) 177

Panel C. Project Return (PR) treatments

male borrowers female borrowers

male lenders 17 (40%) 25 (60%) 42

female lenders 10 (21%) 38 (79%) 48

all lenders 27 (30%) 63 (70%) 90

Page 47: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  47  

Table 8: Predicting lenders’ top 3 borrowers by rank-ordered logit model Dependent Variable: borrower j’s value assigned by lender i

(1) (2) (3) (4) Female borrower 0.40*** 0.41*** 0.37*** 0.34*** (0.073) (0.073) (0.078) (0.096)

Asian borrower 0.19** 0.26*** -0.0323 -0.071 (0.092) (0.093) (0.11) (0.13)

African borrower 0.28*** 0.36*** 0.161 0.033 (0.089) (0.090) (0.10) (0.12)

Farmer borrower 0.83*** 1.05*** 0.43*** 0.41*** (0.076) (0.083) (0.097) (0.12)

Same gender dummy (L=B gender)

0.31*** (0.073)

0.30*** (0.073)

0.27*** (0.074)

0.22** (0.091)

Same ethnicity dummy (L=B ethnicity)

0.27** (0.11)

0.27** (0.11)

0.25** (0.11)

0.25* (0.14)

lenders' standardized perceived riskiness level of the borrower

-0.33*** (0.0430)

-0.65*** (0.052)

-0.72*** (0.064)

lenders' standardized perceived neediness level of the borrower

0.81*** (0.060)

0.70*** (0.071)

average raters' perceived attractiveness level of the borrower (1=not attractive, 7=very attractive)

0.0059 (0.048)

0.098 (0.060)

average raters' perceived happiness level of the borrower (1=not happy, 7=very happy)

0.028 (0.032)

-0.014 (0.040)

average raters' perceived weight of the borrower (1=underweight, 7=overweight)

-0.11** (0.046)

-0.063 (0.056)

PR*Female borrower 0.084 (0.17)

PR*Asian borrower 0.10 (0.25)

Page 48: Altruism Identity and Financial Returns 2-21-2015josiechen.weebly.com/uploads/4/0/2/5/40252223/chen...3 Kiva.org is perhaps the largest microfinance websites in the world, with about

  48  

PR*African borrower 0.41* (0.22)

PR*Farmer borrower 0.042 (0.22)

PR*Same gender dummy (L=B gender)

0.16 (0.16)

PR*Same ethnicity dummy (L=B ethnicity)

-0.049 (0.24)

PR*lenders' standardized perceived riskiness level of the borrower

0.17 (0.11)

PR*lenders' standardized perceived neediness level of the borrower

0.39*** (0.14)

PR*average raters' perceived attractiveness level of the borrower

-0.28*** (0.11)

PR*average raters' perceived happiness level of the borrower

0.10 (0.066)

PR*average raters' perceived weight of the borrower

-0.16 (0.10)

Observations 3,204 3,204 3,204 3,204 Number of groups 267 267 267 267 Log likelihood -1825 -1795 -1690 -1662 LR chi2 188 248 457 513 Prob > chi2 <0.0001 <0.0001 <0.0001 <0.0001 Note: Below we listed the average value (standard deviation) of each control variables: the average value (standard deviation) for “ lenders' standardized perceived neediness level regarding the borrower” is 0 (0.96), for “ lenders' standardized perceived riskiness level regarding the borrower” is 0 (0.96), for “ average raters' standardized perceived attractiveness of the borrower” is 3.55 (0.92), for “ average raters' standardized perceived happiness of the borrower” is 3.82 (1.43), for “ average raters' standardized perceived weight of the borrower” is 3.94 (1.04).