Download as pdf or txt
Download as pdf or txt
You are on page 1of 47

Donation to Minorities: Is the Upper Class Less Susceptible to Empathetic

Charitable Nudge?

Jie Chen1

Bachir Kassas2

Rodolfo M. Nayga Jr.3

1
Graduate Student, Food and Resource Economics Department, University of Florida.
1084 McCarty Hall B, Gainesville, FL, 32611. Email: chenj5@ufl.edu.

2
Corresponding Author. Assistant Professor, Food and Resource Economics
Department, University of Florida. 1099 McCarty Hall B, Gainesville, FL, 32611.
Email: b.kassas@ufl.edu. Phone: (352) 294-7621.

3
Professor and Department Head, Agricultural Economics Department, Texas A&M
University. 309 AGLS building, 2124 TAMU, College Station, TX, 77843. Email:
rudy.nayga@ag.tamu.edu

1
Abstract

Systemic racism threatens access to equal opportunity, resulting in loss of social


welfare for racial/ethnic minorities. Charitable campaigns are a common strategy for
addressing this issue by securing more resources for minority groups. To improve the
effectiveness of charitable campaigns, empathetic nudges are often employed to
encourage higher donations by inducing potential donors to merge themselves into the
victims’ feelings and emotions. While previous studies have investigated the
effectiveness of empathetic nudges, hypothetical settings and lack of sample
representation have limited the generalizability of their findings. We conducted an
online, incentive-compatible experiment using a nationally representative sample of
US adults to examine the effectiveness of three empathetic nudging treatments that
vary in salience by having subjects either read, reflect, or relate to the content of the
nudge. Our results demonstrate the effectiveness of empathetic nudges for the full
sample, where subjects in all three treatments donated significantly more than the
control. Analysis of treatment effect heterogeneity revealed an interesting asymmetry,
where lower-income class individuals responded more spontaneously and effortlessly
to minorities’ struggles (spontaneous involvement), while upper-income class
individuals only responded when they were forced to engage with such narratives
(forced involvement). These results support the idea that upper-class individuals are
less swayed by others’ hardship and thus less susceptible to empathetic nudges.

Keywords: Racial Injustice; Charitable Giving; Empathetic Nudging; Social Classes


JEL Codes: B55; C91; D91

2
1. Introduction

Systemic racism is deeply embedded in US society. Racial prejudice threatens the

livelihoods of racial/ethnic minorities across several dimensions, including health care

resources (Muncan 2018; Nelson 2002; Do and Reanne, 2021; Fiscella and Mechelle,

2016), academic outcomes (Seo et al., 2019; Bhopal and Thandeka, 2019, Kogachi

and Sandra, 2020; Massey and Mary, 2006; Verkuyten et al., 2019), career satisfaction

(Niemann and John, 1998), neighborhood safety (Sampson et al., 2005; Ulmer et al.,

2012; Zimmerman and Steven, 2013), and criminal justice (Crutchfield et al., 2010;

Sampson and Janet, 1997). Consequently, government and nonprofit organizations

have taken various initiatives – such as education, listening programs, and street

protests – to fight against racial inequality. Among these, one common way is to

initiate charitable donation programs to promote equal opportunities for racial/ethnic

minorities (e.g., Black Live Matter).

This study aims to examine the effectiveness of empathetic nudging, a popular

persuasion scenario, in eliciting public donations to ethnic minorities. Three

empathetic nudging treatments are investigated to gain better understanding of the

mechanism through which empathetic nudging influences individual charitable giving

behavior. This will provide critical knowledge on how to optimize empathetic

nudging campaigns to promote higher donation levels that can improve the welfare of

individuals from underrepresented minority groups.

Previous research has shown that the success of fundraising campaigns largely

depends on donors’ pure altruism, which is an internal motivation that cannot be

3
easily swayed by contextual factors (Andreoni, 1990; Ottoni-Wilhelm et al., 2017;

Batson 2010). The empathetic nudging strategy has been pervasively employed in

practice to appeal to altruism and improve persuasion efficiency, mainly through

narratives that depict the misfortune or hardship victims are suffering (Hung and

Robert, 2009; 2004). The empathetic nudging narratives presume that exposure to

others’ hardships/misfortune would induce empathy and help donors better internalize

with, even merge themselves to, victims’ feelings and emotions (Escalas 2007; 2004;

Green and Timothy, 2000; Batson et al., 1997; 2003).

Indeed, a long stream of research shows that individuals with a higher level of

empathy exhibit more prosociality (Smith et al., 2020; Cohen and Cynthia, 2013;

Verhaert and Dirk,2011; Ein‐Gar and Liat, 2013). Empathy essentially reflects

individuals’ affective ability to involuntarily merge into others’ feelings and emotions

with the loss of personal standpoint (Escalas and Barbara, 2003). Although some

evidence supports the effectiveness of providing victims’ stories, and asking

individuals to take others’ perspectives when making charitable donations (Hung and

Robert, 2009; 2004; Batson et al., 2003; 1997; 1999; Sheng and Shihui, 2012),

previous studies suffer from two main limitations.

One concern relates to the hypothetical nature of previous studies and to the

use of student samples (Hung and Robert, 2009; 2004). On the one hand, hypothetical

donations can lack proper incentives, which decreases participants’ involvement and

undermines the accuracy of the elicited prosocial preferences, since subjects do not

face real monetary costs and might therefore exhibit more socially desirable behavior

4
than they would in an incentive-compatible setting. On the other hand, student

samples lack representation of the general population, which imposes challenges on

the generalizability of the results and applicability of implications to policymakers.

Second, previous studies examined the heterogeneous effects of empathetic

nudging mainly using psychological measures, rather than observable characteristics,

which limits their applicability in persuasion content design. The extant literature

pointed out that the effectiveness of directly asking participants to take others’

perspectives depends on the individual’s ability to perform mental simulation (Wyer

et al., 2002; Wyer and Gabriel 1999; Hung and Robert, 2004) and self-focused

attention (Hung and Robert, 2009). For example, Hung and Robert (2009) showed

that self-focused attention only increases the effectiveness of asking to take victims’

perspective when participants have a high/moderate level of similarity with

protagonists. However, mental simulation and self-focused attention are psychological

traits that are hard to observe, let alone incorporate in actionable initiatives to promote

higher donations. When applying empathetic nudging into practice, it is difficult to

account for differences in these unobservable psychological traits. This often

constrains policymakers and program directors to assume that potential targets have

similar responses to information describing others’ hardship, thereby resorting to

uniform information treatments that might not carry the same efficiency across

individuals, leading to a potential loss of important resources.

We conducted an online incentive-compatible experiment with a nationally

representative sample to address these limitations and to test the effectiveness of

5
various empathetic nudging treatments that vary the salience of exposure to others’

hardships. Subjects were randomly assigned to four conditions (control, information,

reflection, and relatability). Participants in the control condition did not receive any

information, whereas those in the three treatment conditions were asked to read an

anecdote depicting struggles and hardships faced by an African American person

based on the color of his skin. Following the reading task, subjects in the information

treatment went straight to making their donation decisions. On the other hand,

subjects in the reflection treatment were asked to report their feelings if they were in

the victim’s shoes, while those in the relatability treatment were asked to relate to the

victim with a personal experience from their past before making their donations.

Using treatments that vary in the salience of exposure to racial/ethnic

minorities’ hardship enriches related empathetic nudging literature by providing a

comprehensive view of donors’ prosocial preferences. The information treatment

captures participants’ spontaneous involvements with mere information presentation in

the empathetic nudge (Zaki and Jason, 2013). Conversely, the reflection and

relatability treatments capture their forced involvements by investigating the effects of

asking for more effortful and engaging responses to nudges that provide specific

thinking directions for potential donors. To this extent, our study helps shed light on

two distinct mechanisms that potentially underly the effect of empathetic nudging on

prosocial decisions.

Subjects’ donations were elicited in an incentive-compatible framework and

used as the measure of prosocial preferences in this study. Each subject was allocated

6
$10 and asked to decide how much they would like to transfer over to a charity of

their choice, from four charity options that cater to the social welfare of African

Americans. Incentives were induced by randomly selecting 10 participants to receive

$10 minus the amount they chose to transfer to the charity, with the remaining balance

deposited in a GoFundMe account for said charity. Subjects were clearly informed

about this beforehand and were provided with the link to each GoFundMe account.

Our design holds several advantages. First, unlike other related studies, the

incentive-compatible nature induces consequentiality in the subjects’ donation

decisions, thereby more accurately eliciting subjects’ true prosocial preferences.

Second, the use of a nationally representative sample, with information on

socioeconomic and behavioral characteristics, allows us to better generalize and apply

the findings into practice, and to capture potential heterogeneities in treatment effects

across observable characteristics that are readily available to policymakers and

program directors.

Third, by varying the salience of nudging across multiple treatments, we can

investigate the effectiveness of thought exercises to internalize the content of the

empathetic nudge (i.e., reflection and relatability treatments), in addition to the effect

of simply presenting the information (i.e., information treatment). On the one hand, as

noted by Green and Timothy (2000), individuals may spontaneously transport

themselves to the protagonists’ roles if the narratives are emotionally absorbing. On

the other hand, studies also documented the effectiveness of asking individuals to

imagine victims’ feelings in charitable giving (Hung and Robert, 2004; 2009). If the

7
information treatment increases individual charitable giving, we can conclude that

people can spontaneously respond to others’ hardships without any further persuasion

(Green and Timothy, 2000). Otherwise, the effectiveness of the reflection and

relatability treatments suggests that individuals are not fully able to spontaneously

internalize others’ hardships and deeper nudging mechanisms are needed to elicit

prosociality.

Our full sample analysis suggests that all three treatments significantly

increased subjects’ donations compared to the control group, with no significant

differences across treatment effects. This lends support to the notion of spontaneous

involvement, where individuals are able to spontaneously and effortlessly respond to

information about racial injustice. However, sub-analysis using a Latent Class Model

(LCM) revealed significant treatment effect heterogeneities. Specifically, while

individuals classified as lower class were equally responsive to all treatments, upper

class individuals only increased their charitable giving under the reflection and

relatability treatments.1 This suggests that spontaneous giving does not explain the

behavior of all subjects, and that the forced involvement, where individuals were

required to respond to more engaged and effortful nudges, dominates for upper class

individuals. In other words, the upper class is less responsive to empathetic nudging

narratives than their lower-class counterparts.

Our study contributes to the extant charitable giving literature by using a

nationally representative sample in an incentive-compatible experiment to examine

1
The lower class refers to individuals with lower income, less education, and higher unemployment.
8
spontaneous and forced involvement to empathetic nudging mechanisms. Although

some related studies examined the effectiveness of asking to take victims’

perspectives in charitable giving, they did not provide a systematic answer to whether

individuals can spontaneously react to others’ hardships or if further nudging needs to

be included when designing the persuasion content (Hung and Robert, 2004; 2009;

Batson et al., 2003; 1997; 1999). We address this issue by varying the salience of the

empathetic nudge and testing for differential treatment effects. We show that

spontaneous and forced involvement with narratives are both intervened in the

mechanism underlying individuals’ capability to react to empathetic nudging, but that

they work individually and exclusively on different population classes.

Our findings also shed light on the long-standing controversial literature

regarding the links between social class and prosociality. One stream of research

supports the notion that people in a higher social class are more unethical and less

prosocial than their lower-class counterparts (Piff et al., 2012; 2010; Guinote et al.,

2015; Vieites, 2017). However, these findings have been challenged by a set of

replication experiments where researchers find null, even positive, associations

between social class and prosociality (Stamos et al., 2020; Korndörfer et al., 2015;

von Hermanni and Andreas, 2019; Andreoni et al., 2021). Although researchers

examined multiple possible explanations for the inconsistent findings, further

investigation is still needed (Côté et al., 2015; Schmukle et al., 2019; Vieites et al.,

2022; Rao et al., 2022).

Our results indicate that the prosociality gap among social classes may be

9
explained by differences in ability to internalize others’ hardship. First, findings in

this study resonate with previous literature by documenting that the upper class is

innately more prosocial than their lower-class counterparts (i.e., they significantly

donated more in the control condition). Second, the information treatment elicited

significantly more donations from the lower class but not the upper class, which

indicates that upper class individuals are less capable of perceiving and understanding

others’ misfortunes, that is, lack of empathy. In other words, although upper class

individuals originally showed more prosociality, they seem to lack ability to

spontaneously take others’ perspectives and perceive victims’ feelings and emotions

without a more engaged and effortful nudge.

Our findings provide important implications for policymakers to enhance the

success of fundraising programs aimed at improving the social welfare of racial/ethnic

minorities. At the full sample level, we documented the effectiveness of three

treatments (i.e., information, reflection, and relatability) in increasing donations with

a nationally representative sample. More importantly, we differentiated the partial

success of mere information treatments (only on individuals from the lower class)

from the full success of reflection and relatability treatments (for individuals from

both the upper and lower class). The effectiveness of the reflection and relatability

treatments on the full sample suggests that when considering empathetic nudging

narratives, it is important to involve thought processes to help potential donors

internalize the content to improve persuasion efficiency.

2. Experimental Design

10
2.1 Participants

An online experiment was conducted on Qualtrics using a nationally representative

sample of 1,015 US adults (18 years and older). Proportional representation was

achieved based on age, gender, income, and geographic region (Northeast, Midwest,

West, and South). After providing consent, subjects confirmed their willingness to pay

attention to the instructions and to provide truthful answers before starting the

experiment. To further improve data quality, an attention check question was added to

filter out subjects who did not pay attention to the instructions. Two subjects were

excluded from the analysis, one for spending an unreasonable amount of time

(approximately 3 hours) reading the text in the information nudge and the other for

stating a donation of more than $10, resulting in a total sample size of 1,013 subjects.2

2.2. Treatments and Donation Task

Subjects were randomly assigned to a control and three treatment conditions:

information, reflection, and relatability. Unlike the control group, subjects in the three

treatment conditions were presented with an anecdote depicting the inequalities and

hardships faced by an African American individual due to racial injustice. Subjects in

the information treatment were only asked to read the text. One the other hand,

subjects in the reflection treatment were also asked to describe how they would feel if

they were in the victim’s position, while subjects in the relatability treatment were

asked to relate to the victim with a past experience. To increase the salience of the

reflection and relatability treatments, subjects were asked to type their responses,
2
The subject reported a donation amount of $1000. While this might have been a typo, meant to
indicate a donation of $10.00, we decided to exclude this subject since we cannot guarantee their true
intended donation amount.
11
which were checked for quality and validity. Subjects were informed that their

responses are completely anonymous.

Following the treatments, all participants completed a donation task, where

each was allocated $10 and asked to decide how much of this allocation to transfer

over to a charity of their choice from four different charity options that catered to the

social welfare of African Americans: Black Lives Matter, Black Girls Code, Thurgood

Marshal College Fund, NAACP Legal Defense and Education Fund. Incentives were

induced by randomly selecting 10 subjects at the end of the study to receive their $10

allocation minus the amount they chose to transfer to the charity of their choice. To

establish participants’ trust in this incentive process, separate non-profit fundraiser

campaigns were created for each of the four charity options on the GoFundMe

platform. Participants were provided with the links to each fundraiser and were

informed that, if selected, the amount of their donation would be automatically

deposited in the fundraiser of their choice, which would create a public record

showing that their donation to the charity actually occurred. The randomly selected

participants were mailed a gift card with the amount of money that they chose to keep

for themselves, if any. After choosing their donations, subjects were asked to report

how likely they think they will be selected among the 10 participants for real payment.

This was used to control for subjects’ perceived probability of being selected, which

was measured on a scale from 0-100.

2.3 Variables

The amounts of subjects’ donations to the charity options are used as the main

12
outcome variable. Sociodemographic and behavioral characteristics were also

collected and used as control variables to check the robustness of the results and to

explore potential treatment effect heterogeneity. Sociodemographic variables included

age, income, household size, and education (coded as categorical variables), as well as

gender, political orientation, race, and employment (coded as binary variables). Prior

literature has demonstrated significant effects of those demographic and

socioeconomic factors on individual prosocial behaviors (Kamas and Anne, 2021;

Mayr and Alexandra, 2020; Matsumoto et al., 2016; Eagly, 2009; Olson et al., 2021;

Andreoni et al., 2021; Ahmed, 2008; Orom et al., 2021). Thus, it is appropriate to

control for these variables in our analysis.

The behavioral variables related to subjects’ experience with

racism/discrimination and their previous interaction with ethnic minorities were

collected as well. Specifically, we asked participants whether they were subject to any

type of discrimination (i.e., racism, gender, or sexual discrimination) or witnessed acts

of discrimination in the past. They also reported the frequency of their interaction

with African Americans and other ethnic minorities, as well as whether they had

befriended an African American/ethnic minority. We also asked subjects whether they

liked any food, music, or celebrities originating from ethnic minority groups. Finally,

subjects also indicated their degree of spirituality and whether they followed any

particular religion. Extant literature has documented the positive effects of perceived

similarities and psychological closeness to victims on donors’ responses to empathetic

charitable appeals (Small and Uri, 2008; Meyer et al., 2012; Hung and Robert, 2009),

13
which highlights the importance of controlling for the behavioral factors described

above.

3. Results

3.1 Sample characteristics

A summary of subject characteristics is reported in Table 1. Our sample consists of

approximately 52.2% males and 74.8% White/Caucasian participants. The median age

is between 45 and 54 years and the average household size is 3. Around half the

participants are either full or part-time employed and received at least 4-year college

education. In addition, average giving to charity is around $4.84, out of the $10

endowment, and the average perceived probability of being selected for real payment

is 38.88%, suggesting the effectiveness of our incentive-compatible design.

3.2 Treatment Balance Check

A breakdown of subject characteristics by treatment is provided in Table 2, along with

the sample size for each treatment. The control and information treatment contained

261 and 262 subjects, respectively, with 249 subjects in the reflection treatment, and

the remaining 241 subjects in the relatability treatment. A non-parametric

Kruskal-Wallis test shows no significant difference between the four conditions in any

of the sociodemographic control variables. We also report the average time subjects

spent reading the racism anecdote, which was not statistically significant across

treatment conditions, suggesting equal attention devoted to the content of the

14
treatment under each condition (information, reflection, relatability). The results in

Table 2 suggest a balanced sample that supports investigation of treatment effects on

donation decisions.

3.3 Analyzing Main Treatment Effect

Average charitable giving by treatment is presented in Figure 1 and Table 3. Results

of non-parametric pairwise comparison tests indicate that subjects in the information,

reflection, and relatability treatments donated significantly more than those in the

control group (Mann-Whitney tests, p-value=0.024, 0.000, and 0.006, respectively).

On the other hand, we did not find significant differences in donations across the three

treatment conditions (Mann-Whitney tests, p-value>0.141).3 This indicates that more

engaging empathetic nudges (i.e., reflection and relatability treatments) do not

significantly outperform the mere presentation of information (i.e., information

treatment) in eliciting donations to promote racial/ethnic minorities’ equal opportunity.

Our full sample analysis therefore lends support to the notion of spontaneous

involvement in response to empathetic nudging narratives. This leads to the first

result.

Result 1: The general effect of the information, reflection, and relatability treatments

lends support to spontaneous involvement with minorities’ misfortune through equal

effectiveness of all treatments in increasing subjects’ donations.

Similar results are illustrated in Figures 2 and 3, which plot the histogram and

cumulative distribution function of donations, respectively, across treatments. In the

3
p-value=0.141 (information vs. reflection), 0.482 (information vs. relatability), 0.467 (reflection vs.
relatability).
15
control, 32.95% of subjects chose to keep the entire $10 allocation, which is higher

than the information (27.1%), reflection (20.08%), and relatability (28.22%)

treatments. In addition, subjects in the control group had the lowest fraction who

contributed the entire allocation, with only 23.75%, compared to 30.53% in the

information, 36.55% in the reflection, and 36.93% in the relatability treatment. A

Kolmogorov–Smirnov test was conducted to examine differences in the distribution

of giving between the control and each of the three treatments. The results were

significant for the reflection (p-value=0.011) and relatability (p-value=0.024)

treatments, but not for the information treatment (p-value=0.111).

3.4 Analyzing Treatment Effect Heterogeneity

We conducted a latent class analysis (LCA) to examine treatment effect heterogeneity

across a set of observable sociodemographic and behavioral variables that are

positively correlated with prosocial preferences (Kamas and Anne, 2021; Mayr and

Alexandra, 2020; Matsumoto et al., 2016).4 We included all sociodemographic

variables as well as past discrimination experience into the LCA model. This selection

was based on prior literature and additional analysis (reported in Table 5 in the

robustness check subsection), both of which showed significance for most of these

variables.5 The LCA model allows the classification of subjects into different types,

based on underlying observable characteristics. A detailed review of this method is

provided by Collins and Stephanie (2009). The LCA model only allowed us to

4
We also created interaction terms for all significant variables in Table 5. However, the Tobit
regression results did not show a strong interacting effect (please see Table A3), which led us to
conduct a follow-up latent class analysis.
5
LCA Results including only sociodemographic characteristics are reported in Table A1 in the appendix and are
consistent with our findings.
16
classify our sample into two groups, as models with higher number of classifications

failed to converge.

The LCA model results are presented in Table 4. Approximately 64.46% of the

full sample were classified into the first class and the remaining 35.34% were

classified into the second class. Class 2 is characterized by a higher fraction of

individuals who are male, employed, Caucasian, married, and have a higher income

and educational attainment, indicating a higher social status for these individuals

compared to class 1. Based on this classification, we labeled individuals in class 1 and

class 2 as the lower and upper class, respectively.

Figure 4 presents a breakdown of average giving across treatment and latent

class. The lower class significantly gave more in the information, reflection, and

relatability treatments compared to the control group (Mann-Whitney test,

p-value=0.021, 0.001, and 0.034, respectively). Conversely, we only find suggestive

evidence of higher donations from the upper class under the reflection

(Mann-Whitney test, p-value=0.057) and relatability (Mann-Whitney test,

p-value=0.088) treatments, but not the information treatment (Mann-Whitney test,

p-value=0.548), compared to the control group. Comparing donations between the

upper and lower class, we find significantly higher giving from the upper class under

the control condition (Mann-Whitney test, p-value=0.015). These results indicates that

while the upper class initially show higher prosocial tendencies, they are less

responsive to empathetic nudging narratives as they require more engaging and

effortful nudging to increase their donation levels further. On the other hand, the

17
lower class seems able to better internalize the content of empathetic nudging

narratives more spontaneously and effortlessly given that they significantly responded

to the mere presentation of others’ hardship (i.e., information treatment) by increasing

their donation levels. This leads to our second result.

Result 2: The LCA model suggests a duality in the mechanism underlying

responsiveness to empathetic nudging, where spontaneous involvement with others’

misfortunes is evident among the lower class, but forced involvement dominates the

responsiveness of the upper class.

3.5 Robustness Check

We check the robustness of our results using two approaches. First, multiple Tobit

regression specifications, with a set of controls, were estimated to test the treatment

effects on subjects’ donations. The Tobit model was used to account for the clustering

of donations at the lower ($0) and upper ($10) bounds. Results using the full sample

are presented in Table 5. The first specification is a baseline model that only includes

treatment indicators as explanatory variables. Sociodemographic and behavioral

variables are controlled for in the second and third specifications, respectively. The

fourth specification is a saturated model that includes all covariates. Notably, the

results only provide suggestive evidence on the effectiveness of the information

treatment, which was significant only in specifications [1] and [3], but we find robust

and consistent evidence of higher donations under the reflection and relatability

treatments in all specifications. The lack of consistency in the significance of the

information treatment might be a result of this treatment being only effective on the

18
lower class, as was shown earlier.

Looking at the control variables, our results echo the extant literature that

female, age, educational attainment, democratic affiliation, and prior discrimination

experience are positively correlated with charitable giving to enhance racial/ethnic

minorities’ welfare (Kamas and Anne, 2021; Mayr and Alexandra, 2020; Matsumoto

et al., 2016; Eagly, 2009; Olson et al., 2021; Andreoni et al., 2021; Ahmed, 2008;

Orom et al., 2021).

Tobit regressions were also estimated separately for subjects in the lower and

upper class. Two specifications were estimated for each class, one as a baseline model

and the other controlling for behavioral factors. Sociodemographic variables were not

included in these models since they were already captured in the LCA model to

separate subjects into the two classes. The results are presented in Table 6 and

conform to the previous findings surrounding the treatment effects on donations of the

lower and upper class. Specifically, the lower class donated more under all three

treatments, while suggestive evidence was found for higher donations among the

upper class only under the reflection and relatability treatments. These results were

robust across specifications.

The second approach to check the robustness of our findings is by coding

subjects’ donations into 5 categories: selfish ( �������� = $0 ), self-leaning ( $0 <

�������� < $5) , balanced ( �������� = $5 ), selfless-leaning ( $5 < �������� <

$10), and selfless (�������� = $10). This resulted in a 5-level categorical variable,

taking higher values for higher prosociality in donations. A breakdown of donations

19
across the five categories is provided for each treatment in Figure 5 and Table 7.

Cross-treatment comparisons revealed significant differences in the fraction of selfish,

self-leaning, and selfless donations. Consistent with the previous findings, the control

group contained the highest fraction of selfish and self-leaning donations, and the

lowest fraction of selfless donations.

The categorical variable describing donation type was also used in a set of

ordered Logit regressions estimating the treatment effects on subjects’ donation levels.

The results using the full sample are reported in Table 8, while individual regressions

on the subsamples of lower and upper class are reported in Table 9. The same

specifications from the Tobit model were estimated in the ordered Logit regressions,

and the results are very similar across the two approaches. At the full sample level, we

again find consistent evidence of higher donation levels under the reflection and

relatability treatments for all specifications, but inconsistent evidence for the

information treatment, where the coefficient was not statistically significant for

specification [2]. As mentioned before, this could be due to the fact that the

information treatment was only effective for individuals in the lower class. Looking at

the separate results for lower and upper class, we again see higher donation levels

under all treatments for the former and higher donation levels only under the

reflection and relatability treatment for the latter. Together, the Tobit and ordered

Logit analyses provide support for the robustness of the results presented in this study.

4. Discussion and Conclusion

Systemic racism, which is deeply embedded in US society, significantly threatens

20
racial/ethnic minorities’ access to equal opportunity across multiple dimensions.

Governments and non-profit organizations have initiated numerous charitable giving

campaigns to alleviate this social issue and improve the welfare of minority groups. In

doing so, a popular strategy to elicit higher donations relies on empathetic nudging,

where narratives are provided that depict the struggles and hardships of ethnic

minorities in an attempt to induce empathetic feelings by helping donors internalize

and merge themselves with the victims’ feelings and emotions (Hung and Robert,

2009; 2004; Escalas 2007; 2004; Green and Timothy, 2000; Batson et al., 1997; 2003).

Despite their suggested effectiveness, an understanding of the mechanism underlying

individuals’ responsiveness to empathetic nudges is still lacking.

We conducted a nationally representative online experiment to test the

effectiveness of empathetic nudging narratives under multiple treatments (information,

reflection, relatability) that vary the salience of the information presented in the nudge.

This allowed us to examine participants’ spontaneous vs. forced involvement to the

empathetic narrative by comparing subjects’ charitable donations across treatments.

Results of full sample analysis show that all three treatments were equally

effective in eliciting significantly more donations than the control, supporting the idea

of spontaneous involvement with minorities’ struggles. Donation levels were not

significantly different between the information, reflection, and relatability treatments,

indicating that forced involvement does not provide any additional benefits over

spontaneous prosociality on a full sample level.

We also highlight significant heterogeneity in treatment effects between lower

21
and upper-class individuals using a LCA model. Specifically, the lower class were

equally responsive to all three treatments, while the upper class only showed

significant increases in their donations under the reflection and relatability treatments.

This suggests an asymmetric mechanism driving individuals’ responsiveness to

empathetic nudging, where spontaneous involvement to empathetic narratives is more

prevalent for the lower class, while forced involvement dominates the behavior of the

upper class.

Our findings contribute to fundraising literature in multiple ways. First, we

shed light on the behavioral mechanism underlying responsiveness to empathetic

nudging, which provides useful information for improving the effectiveness of this

type of nudge. We show that ability to spontaneously and effortlessly internalize the

content of empathetic nudges is not general, and only exists among a subsample of the

population (the lower class). On the other hand, a sizeable fraction of individuals (the

upper class) requires more engagement with the nudge, through thought processes to

help internalize the information, before showing significant changes in donation

behavior.

Second, the difference in donation behavior between lower- and upper-class

individuals uncovered in this study can help address the controversial findings in the

literature regarding the link between social class and prosociality. We show that while

the upper class are generally more prosocial (donating higher amounts in the control

condition), they are also less able to internalize and merge themselves with the

victims’ perspective in empathetic nudges that are not sufficiently engaging. This

22
conforms to previous evidence that upper class individuals are generally less

empathetic than lower class individuals, where they are less able to predict others’

emotions and activate fewer neural responses to others’ pain (Varnum et al., 2015;

Kraus et al., 2010).

Third, our analysis of treatment effect heterogeneity can help inform

policymakers and non-profit organizations on the types of strategies in empathetic

nudges that are more likely to elicit higher responsiveness levels. Specifically, we

highlight the partial success of mere information presentation in empathetic nudges,

which only works on a subset of individuals, compared to the full success of inducing

thought processes to internalize the content of the nudge (i.e., reflection and

relatability treatments). This result suggests that when considering empathetic nudges,

policymakers and fundraising organizations should include procedures that induce

individuals to internalize the hardships and loss of social welfare experienced by

victims to improve the effectiveness of their nudges.

Our study has some limitations, which provide directions for future research.

For instance, although our results closely echo previous findings that upper class

individuals are less capable of perceiving and empathizing with others’ feelings and

emotions compared to their lower-class peers (Varnum et al., 2015; Kraus et al., 2010),

we do not directly measure subjects’ objective empathetic feelings in this study. One

useful tool for obtaining such objective measures of empathy is the use of biometric

equipment to collect psychophysiological data (e.g., facial expressions of emotion,

23
pupil dilation, electrical brain activity) that signal subjects’ feelings and emotions in

response to the empathetic nudging narrative.

Reference
Andreoni, James. "Impure altruism and donations to public goods: A theory of
warm-glow giving." The economic journal 100, no. 401 (1990): 464-477.
Andreoni, James, Nikos Nikiforakis, and Jan Stoop. "Higher socioeconomic status
does not predict decreased prosocial behavior in a field experiment." Nature
communications 12, no. 1 (2021): 1-8.
Batson, C. Daniel. "Empathy-induced altruistic motivation." (2010).
Batson, C. Daniel, David A. Lishner, Amy Carpenter, Luis Dulin, Sanna
Harjusola-Webb, Eric L. Stocks, Shawna Gale, Omar Hassan, and Brenda Sampat.
"“... As you would have them do unto you”: Does imagining yourself in the other's
place stimulate moral action?." Personality and social psychology bulletin 29, no.
9 (2003): 1190-1201.
Batson, C. Daniel, Karen Sager, Eric Garst, Misook Kang, Kostia Rubchinsky, and
Karen Dawson. "Is empathy-induced helping due to self–other merging?." Journal
of personality and social psychology 73, no. 3 (1997): 495.
24
Batson, C. Daniel, Nadia Ahmad, Jodi Yin, Steven J. Bedell, Jennifer W. Johnson, and
Christie M. Templin. "Two threats to the common good: Self-interested egoism
and empathy-induced altruism." Personality and Social Psychology Bulletin 25,
no. 1 (1999): 3-16.
Bhopal, Kalwant, and Thandeka K. Chapman. "International minority ethnic
academics at predominantly white institutions." British Journal of Sociology of
Education 40, no. 1 (2019): 98-113.
Christov-Moore, Leonardo, Elizabeth A. Simpson, Gino Coudé, Kristina Grigaityte,
Marco Iacoboni, and Pier Francesco Ferrari. "Empathy: Gender effects in brain
and behavior." Neuroscience & biobehavioral reviews 46 (2014): 604-627.
Cohen, Elizabeth L., and Cynthia Hoffner. "Gifts of giving: The role of empathy and
perceived benefits to others and self in young adults’ decisions to become organ
donors." Journal of Health Psychology 18, no. 1 (2013): 128-138.
Collins, Linda M., and Stephanie T. Lanza. Latent class and latent transition analysis:
with applications in the social, behavioral, and health sciences. Vol. 718. John
Wiley & Sons, 2009.
Côté, Stéphane, Julian House, and Robb Willer. "High economic inequality leads
higher-income individuals to be less generous." Proceedings of the National
Academy of Sciences 112, no. 52 (2015): 15838-15843.
Crutchfield, Robert D., April Fernandes, and Jorge Martinez. "Racial and ethnic
disparity and criminal justice: How much is too much?." The Journal of Criminal
Law and Criminology (1973-) 100, no. 3 (2010): 903-932.
Do, D. Phuong, and Reanne Frank. "Unequal burdens: assessing the determinants of
elevated COVID-19 case and death rates in New York City’s racial/ethnic
minority neighbourhoods." Journal of Epidemiol Community Health 75, no. 4
(2021): 321-326.
Ein‐Gar, Danit, and Liat Levontin. "Giving from a distance: Putting the charitable
organization at the center of the donation appeal." Journal of Consumer
Psychology 23, no. 2 (2013): 197-211.
Escalas, Jennifer Edson. "Self-referencing and persuasion: Narrative transportation
versus analytical elaboration." Journal of Consumer Research 33, no. 4 (2007):
421-429.
Escalas, Jennifer Edson, and Barbara B. Stern. "Sympathy and empathy: Emotional
responses to advertising dramas." Journal of Consumer Research 29, no. 4 (2003):
566-578.
Fiscella, Kevin, and Mechelle R. Sanders. "Racial and ethnic disparities in the quality
of health care." Annual review of public health 37 (2016): 375-394.
Green, Melanie C., and Timothy C. Brock. "The role of transportation in the
persuasiveness of public narratives." Journal of personality and social psychology
79, no. 5 (2000): 701.
Guinote, Ana, Ioanna Cotzia, Sanpreet Sandhu, and Pramila Siwa. "Social status
modulates prosocial behavior and egalitarianism in preschool children and adults."
Proceedings of the National Academy of Sciences 112, no. 3 (2015): 731-736.
Han, Shihui, Yan Fan, and Lihua Mao. "Gender difference in empathy for pain: an
electrophysiological investigation." Brain research 1196 (2008): 85-93.
25
Hung, Iris W., and Robert S. Wyer Jr. "Differences in perspective and the influence of
charitable appeals: When imagining oneself as the victim is not beneficial."
Journal of Marketing Research 46, no. 3 (2009): 421-434.
Kamas, Linda, and Anne Preston. "Empathy, gender, and prosocial behavior." Journal
of Behavioral and Experimental Economics 92 (2021): 101654.
Kogachi, Kara, and Sandra Graham. "Numerical minority status in middle school and
racial/ethnic segregation in academic classes." Child development 91, no. 6
(2020): 2083-2102.
Korndörfer, Martin, Boris Egloff, and Stefan C. Schmukle. "A large scale test of the
effect of social class on prosocial behavior." PloS one 10, no. 7 (2015): e0133193.
Kraus, Michael W., Stéphane Côté, and Dacher Keltner. "Social class, contextualism,
and empathic accuracy." Psychological science 21, no. 11 (2010): 1716-1723.
Massey, Douglas S., and Mary J. Fischer. "The effect of childhood segregation on
minority academic performance at selective colleges." Ethnic and Racial Studies
29, no. 1 (2006): 1-26.
Meyer, Meghan L., Carrie L. Masten, Yina Ma, Chenbo Wang, Zhenhao Shi, Naomi I.
Eisenberger, and Shihui Han. "Empathy for the social suffering of friends and
strangers recruits distinct patterns of brain activation." Social cognitive and
affective neuroscience 8, no. 4 (2013): 446-454.
Muncan, Brandon. "Cardiovascular disease in racial/ethnic minority populations:
illness burden and overview of community-based interventions." Public Health
Reviews 39, no. 1 (2018): 1-11.
Nelson, Alan. "Unequal treatment: confronting racial and ethnic disparities in health
care." Journal of the national medical association 94, no. 8 (2002): 666.
Niemann, Yolanda Flores, and John F. Dovidio. "Relationship of solo status, academic
rank, and perceived distinctiveness to job satisfaction of racial/ethnic minorities."
Journal of Applied Psychology 83, no. 1 (1998): 55.
Ottoni-Wilhelm, Mark, Lise Vesterlund, and Huan Xie. "Why do people give? Testing
pure and impure altruism." American Economic Review 107, no. 11 (2017):
3617-33.
Piff, Paul K., Daniel M. Stancato, Stéphane Côté, Rodolfo Mendoza-Denton, and
Dacher Keltner. "Higher social class predicts increased unethical behavior."
Proceedings of the National Academy of Sciences 109, no. 11 (2012): 4086-4091.
Piff, Paul K., Michael W. Kraus, Stéphane Côté, Bonnie Hayden Cheng, and Dacher
Keltner. "Having less, giving more: the influence of social class on prosocial
behavior." Journal of personality and social psychology 99, no. 5 (2010): 771.
Rao, Ting‐Ting, Shen‐Long Yang, Feng Yu, Bu‐Xiao Xu, and Jia Wei. "Perception of
class mobility moderates the relationship between social class and prosocial
behaviour." Asian Journal of Social Psychology 25, no. 1 (2022): 88-102.
Sampson, Robert J., and Janet L. Lauritsen. "Racial and ethnic disparities in crime
and criminal justice in the United States." Crime and justice 21 (1997): 311-374.
Schmukle, Stefan C., Martin Korndörfer, and Boris Egloff. "No evidence that
economic inequality moderates the effect of income on generosity." Proceedings
of the National Academy of Sciences 116, no. 20 (2019): 9790-9795.
Seo, Eunjin, Yishan Shen, and Edna C. Alfaro. "Adolescents’ beliefs about math
ability and their relations to STEM career attainment: Joint consideration of
26
race/ethnicity and gender." Journal of Youth and Adolescence 48, no. 2 (2019):
306-325.
Sheng, Feng, and Shihui Han. "Manipulations of cognitive strategies and intergroup
relationships reduce the racial bias in empathic neural responses." NeuroImage 61,
no. 4 (2012): 786-797.
Singer, Tania, Ben Seymour, John P. O'Doherty, Klaas E. Stephan, Raymond J. Dolan,
and Chris D. Frith. "Empathic neural responses are modulated by the perceived
fairness of others." Nature 439, no. 7075 (2006): 466-469.
Smith, Karen E., Greg J. Norman, and Jean Decety. "Medical students’ empathy
positively predicts charitable donation behavior." The journal of positive
psychology 15, no. 6 (2020): 734-742.
Small, Deborah A., and Uri Simonsohn. "Friends of victims: Personal experience and
prosocial behavior." Journal of Consumer Research 35, no. 3 (2008): 532-542.
Stamos, Angelos, Florian Lange, Szu-chi Huang, and Siegfried Dewitte. "Having less,
giving more? Two preregistered replications of the relationship between social
class and prosocial behavior." Journal of Research in Personality 84 (2020):
103902.
Ulmer, Jeffery T., Casey T. Harris, and Darrell Steffensmeier. "Racial and ethnic
disparities in structural disadvantage and crime: White, Black, and Hispanic
comparisons." Social science quarterly 93, no. 3 (2012): 799-819.
Varnum, Michael EW, Chris Blais, Ryan S. Hampton, and Gene A. Brewer. "Social
class affects neural empathic responses." Culture and Brain 3, no. 2 (2015):
122-130.
Verhaert, Griet A., and Dirk Van den Poel. "Empathy as added value in predicting
donation behavior." Journal of Business Research 64, no. 12 (2011): 1288-1295.
Verkuyten, Maykel, Jochem Thijs, and Nadya Gharaei. "Discrimination and academic
(dis) engagement of ethnic-racial minority students: A social identity threat
perspective." Social Psychology of Education 22, no. 2 (2019): 267-290.
Vieites, Yan, Rafael Goldszmidt, and Eduardo B. Andrade. "Social Class Shapes
Donation Allocation Preferences." Journal of Consumer Research 48, no. 5 (2022):
775-795.
Vieites, Yan Bernardes. "The Prosocial class: how social class influences prosocial
behavior." PhD diss., 2017.
von Hermanni, Hagen, and Andreas Tutić. "Does economic inequality moderate the
effect of class on prosocial behavior? A large-scale test of a recent hypothesis by
Côté et al." PLoS One 14, no. 8 (2019): e0220723.
Wyer Jr, Robert S., and Gabriel A. Radvansky. "The comprehension and validation of
social information." Psychological review 106, no. 1 (1999): 89.
Wyer Jr, Robert S., Rashmi Adaval, and Stanley J. Colcombe. "Narrative-based
representations of social knowledge: Their construction and use in comprehension,
memory, and judgment." In Advances in experimental social psychology, vol. 34,
pp. 131-197. Academic Press, 2002.
Zaki, Jamil, and Jason P. Mitchell. "Intuitive prosociality." Current Directions in
Psychological Science 22, no. 6 (2013): 466-470.

27
Zimmerman, Gregory M., and Steven F. Messner. "Individual, family background,
and contextual explanations of racial and ethnic disparities in youths’ exposure to
violence." American journal of public health 103, no. 3 (2013): 435-442.

Tables and Figures

Table 1. Summary Statistics of Participants


Variable % Variable %
Male 52.2 1=less than high school 2.5
Female 47.8 2=high school degree or 18.8
equivalent
3=some college 18.2
1=18-24 12.5 4= 2-year degree 9.2
2=25-34 18.2 5= 4-year degree 26.5
3=35-44 17.3 6=Master’s degree 18.7
4=45-54 17.1 7=Doctorate degree 5.7
5=55-64 16.2 8=Other 0.4
6=65-74 16.6
7=75 or more 2.2 Caucasian 74.8
Hispanic 4.1
1=less than $25,000 19.5 African American 10.5
2=$25,000-$49,999 20.5 Other 10.6
3=$50,000-$74,999 17.2
28
4=$75,000-$99,999 12.5 Employed 56.9
5=$100,000-$149,999 15.1 Student 3.7
6=$150,000-$199,999 7.2 Unemployed 14.7
7=$200,000 or more 8.0 Other 24.7

Married 52.8 Discrimination Experienced 61.2


Not Married 47.2 Discrimination Not Experienced 38.8
Democrat 40.9 Republican 35.1

Mean Median Std. Dev.


Household Size 2.952 3 1.571
Average giving to charity 4.837 5 4.120
Probability of receiving $10 payment 38.880 30 34.848

Table 2. Treatment Balancing


Variable Control Information Reflection Relatability p-value
Group Treatment Treatment Treatment KWallis
(n=261) (n=262) (n=249) (n=241)
Male 0.559 0.557 0.476 0.490 0.117
(0.497) (0.498) (0.500) (0.501)
Age 3.510 3.592 3.656 3.838 0.186
(1.672) (1.688) (1.748) (1.730)
Income 3.264 3.340 3.416 3.456 0.529
(1.880) (1.921) (1.904) (1.786)
Education 4.184 4.286 4.140 4.166 0.812
(1.684) (1.658) (1.670) (1.578)
Household Size 3.038 2.908 2.948 2.909 0.857
(1.647) (1.521) (1.591) (1.525)
Caucasian 0.755 0.729 0.768 0.739 0.752
(0.431) (0.445) (0.423) (0.440)
African American 0.103 0.122 0.092 0.100 0.717
(0.305) (0.328) (0.290) (0.300)
29
Hispanic 0.054 0.038 0.032 0.041 0.658
(0.226) (0.192) (0.176) (0.200)
Marriage 0.513 0.511 0.546 0.544 0.783
(0.501) (0.501) (0.499) (0.499)
Employed 0.586 0.584 0.568 0.535 0.643
(0.493) (0.494) (0.496) (0.500)
Discrimination 0.598 0.607 0.586 0.660 0.354
Experience
(0.491) (0.489) (0.493) (0.475)
Democrat 0.395 0.450 0.386 0.402 0.443
(0.490) (0.5498) (0.488) (0.491)
Republican 0.375 0.309 0.373 0.349 0.354
0.485 (0.463) (0.484) (0.478)
Reading Time --- 98.351 131.119 93.476 0.158
--- (89.161) (321.475) 52.615

Table 3. Average Charitable Giving by Treatment


Mean Std. Dev. p-value
Control Group 4.06 3.99 -
Information Treatment 4.86 4.07 0.023
Reflection Treatment 5.34 4.09 0.000
Relatability Treatment 5.13 4.27 0.004

30
Table 4. Description of Classes from Latent Class Analysis Model
Class 1 Class 2
Lower Class Upper Class
Male 0.384 0.769
Employed 0.399 0.873
White/Caucasian 0.705 0.826
African American 0.126 0.063
Hispanic 0.059 0.010
Married 0.325 0.892
Republican 0.293 0.456
Democrat 0.417 0.394
Discrimination Experience 0.603 0.629
Age 3.802 3.363
Household Income 2.307 5.264
Education 3.390 5.633
Household Size 2.609 3.565

31
Class member percentage 64.46% 35.54%

Table 5. Tobit Regression of Treatment Effects on Charitable Giving


Charitable Giving
[1] [2] [3] [4]
Coefficient Coefficient Coefficient Coefficient
(Std. Err.) (Std. Err.) (Std. Err.) (Std. Err.)
Information 1.777** 1.187 1.667** 1.245
(0.854) (0.813) (0.841) (0.812)
Reflection 3.180*** 2.998*** 3.329*** 3.041***
(0.870) (0.830) (0.858) (0.830)
Relatability 2.457*** 2.024** 2.400*** 1.863**
(0.880) (0.841) (0.864) (0.839)
Male -1.732** -1.462**
(0.671) (0.677)
Age 0.687*** 0.820***
(0.225) (0.232)
Household Income 0.098 0.090
(0.213) (0.212)
32
Education 0.557** 0.461*
(0.238) (0.239)
Employed 0.814 0.654
(0.746) (0.747)
Democrat 1.324* 1.334*
(0.767) (0.765)
Republican -2.674*** -2.489***
(0.811) (0.818)
White/Caucasian -0.627 -0.422
(0.987) (0.990)
Hispanic -0.787 -1.178
(1.690) (1.686)
African American -1.747 -1.700
(1.292) (1.318)
Household Size -0.086 -0.125
(0.217) (0.217)
Witnessed Discrimination 0.249 0.208
(0.770) (0.747)
Experienced Discriminated 1.785** 1.639**
(0.773) (0.754)
Interaction African American -0.332 -0.207
(0.605) (0.597)
Interaction Other Ethnicities 0.327 0.365
(0.580) (0.569)
Befriended African American 0.128 0.482
(1.131) (1.105)
Befriended Other Ethnicity 0.205 0.505
(1.206) (1.178)
Likes Ethnic Celebrities 0.977 0.503
(1.208) (1.182)
Likes Ethnic Music 0.596 1.083
(0.928) (0.919)
Likes Ethnic Food 1.259 0.630
(1.142) (1.116)
Follows a Religion 0.415 0.217
(1.355) (1.317)
Spirituality -0.008 0.034
(0.313) (0.312)
Chance of Receiving Money 0.056*** 0.047*** 0.055***
(0.010) (0.009) (0.010)
Constant 3.289*** -2.017 -3.061 -6.964***
(0.605) (1.713) (2.269) (2.660)

Observations 1,013 1,013 1,013 1,013


33
Notes: The model was estimated with robust standard error to relieve heteroscedasticity bias. *, **, ***
denote p value is less than 0.1, 0.05, 0.01, respectively.

Table 6. Tobit Regression of Treatment Effects on Charitable Giving by Social Class


Lower Class Upper Class
[1] [2] [3] [4]
Coefficient Coefficient Coefficient Coefficient
(Std. Err.) (Std. Err.) (Std. Err.) (Std. Err.)
Information 2.764** 2.294* 0.627 0.967
(1.342) (1.318) (0.997) (0.993)
Reflection 4.321*** 3.971*** 1.971* 1.740*
(1.353) (1.324) (1.037) (1.033)
Relatability 3.057** 2.870** 1.766* 2.010**
(1.380) (1.349) (1.029) (1.016)
Constant 1.990** -4.381 4.786*** -4.097
(0.958) (3.375) (0.702) (3.311)

Controls No Yes No Yes


Observations 653 653 360 360
Log Likelihood -1219.939 -1198.848 -800.047 -791.098
34
Notes: The model was estimated with robust standard error to relieve heteroscedasticity bias. *, **, ***
denote p value is less than 0.1, 0.05, 0.01, respectively.

Table 7. Breakdown of Subject Donation Types by Treatment


Control Information Reflection Relatability
p-value
Group Treatment Treatment Treatment
Selfish 0.330 0.271 0.200 0.282 0.012
(0.471) (0.445) (0.401) (0.451)
Self-Leaning 0.222 0.176 0.216 0.137 0.054
(0.417) (0.381) (0.412) (0.344)
Balanced 0.161 0.187 0.156 0.174 0.785
(0.368) (0.391) (0.364) (0.380)
Selfless-Leaning 0.050 0.061 0.060 0.037 0.610
(0.218) (0.240) (0.238) (0.190)
Selfless 0.238 0.305 0.368 0.369 0.003
(0.426) (0.461) (0.483) (0.484)

35
Table 8. Ordered Logit Regression of Treatment Effects on Charitable Giving
[1] [2] [3] [4]
Coefficient Coefficient Coefficient Coefficient
(Std. Err.) (Std. Err.) (Std. Err.) (Std. Err.)
Information 0.316** 0.258 0.337** 0.272*
(0.157) (0.159) (0.158) (0.161)
Reflection 0.621*** 0.589*** 0.647*** 0.609***
(0.159) (0.161) (0.161) (0.164)
Relatability 0.484*** 0.414** 0.467*** 0.386**
(0.163) (0.166) (0.165) (0.168)
Chance of Receiving Money 0.009*** 0.011*** 0.009*** 0.011***
(0.002) (0.002) (0.002) (0.002)
/cut1 -0.295** 0.300 0.499 1.234**
(0.133) (0.332) (0.427) (0.518)
/cut2 0.569*** 1.204*** 1.373*** 2.147***
(0.134) (0.334) (0.429) (0.520)
/cut3 1.279*** 1.954*** 2.093*** 2.907***
36
(0.139) (0.338) (0.432) (0.524)
/cut4 1.514*** 2.204*** 2.332*** 3.162***
(0.140) (0.339) (0.433) (0.525)
Socioeconomic Controls NO YES NO YES
Behavioral Controls NO NO YES YES

Observations 1,013 1,013 1,013 1,013


Log Likelihood -1484.46 -1447.51 -1475.24 -1438.39

Table 9. Ordered Logit Regression of Treatment Effects on Charitable Giving by


Social Class
Lower Class Upper Class

Coefficient Coefficient Coefficient Coefficient


(Std. Err.) (Std. Err.) (Std. Err.) (Std. Err.)
Information 0.435** 0.405** 0.119 0.243
(0.195) (0.198) (0.264) (0.271)
Reflection 0.653*** 0.649*** 0.537** 0.495*
(0.197) (0.203) (0.260) (0.276)
Relatability 0.480** 0.467** 0.495* 0.578**
(0.215) (0.219) (0.269) (0.280)
/cut1 0.070 0.503 -1.248*** 1.044
(0.165) (0.497) (0.285) (0.869)
/cut2 0.790*** 1.230** -0.015 2.319***
(0.168) (0.498) (0.263) (0.873)
/cut3 1.439*** 1.883*** 0.816*** 3.179***
37
(0.175) (0.500) (0.264) (0.881)
/cut4 1.558*** 2.003*** 1.264*** 3.642***
(0.178) (0.500) (0.268) (0.880)

Controls NO YES NO YES


Observations 653 653 360 360
Log Likelihood -897.298 - -892.785 -556.021 -546.873

Figure 1. Average Charitable Giving by Treatment

38
Figure 2. Distribution of Charitable Giving by Treatment

39
Figure 3. CDF for Charitable Giving by Treatment
Notes: Kolmogorov Smirnov test for difference in distribution. Results: Control Vs. Info
(p-value=0.111); Control Vs. Reflection (p-value=0.011); Control Vs. Relatability (p-value=0.024)

40
Figure 4. Average Charitable Giving by Treatment and Social Class

41
Figure 5. Comparing Subject Donation Types by Treatment

42
Appendix

Table A1. LCA model without discrimination experience


Class 1 Class 2
Lower Class Upper Class
Male 0.384 0.770
Employed 0.399 0.873
White 0.705 0.826
Black 0.126 0.063
Hispanic 0.059 0.010
Married 0.325 0.893
Republican 0.293 0.456
Democrat 0.417 0.394
Age 3.801 3.363
Income 2.308 5.266
Education 3.391 5.634
Household Size 2.610 3.565

Class Percentage (%) 64.57 35.43

43
Table A2. Regression results without discrimination experience
(1) (2) (3) (4)
Lower Class Lower Class Upper Class Upper Class
Info 2.789** 2.303* 0.585 0.870
(1.336) (1.311) (1.002) (0.993)
Reflect 4.311*** 3.958*** 1.973* 1.681
(1.349) (1.319) (1.039) (1.032)
Relate 3.051** 2.862** 1.769* 2.011**
(1.376) (1.345) (1.032) (1.015)
Constant 1.995** -4.273 4.786*** -5.728
(0.955) (3.329) (0.703) (3.496)

Controls NO YES NO YES


Observations 654 654 359 359
Note: Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1

44
Table A3. The regression results with full interaction terms across all significant
controls
Charitable Giving
Coeff.
(Std. Err.)
Info -1.671
(3.365)
Reflect 1.156
(3.224)
Relate -1.055
(3.509)
Male -1.396
(1.285)
Age 0.157
(0.394)
Income 0.009
(0.222)
Education 0.462*
(0.239)
Employed 0.879
(1.360)
Democrat 3.430**
(1.535)
Republican -1.333
(1.570)
White/Caucasian -0.567
(0.990)
Hispanic -1.075
(1.684)
Black/African American -1.824
(1.317)
Household Size -0.176
(0.221)
Witnessed Discrimination 0.136
(0.749)
Discrimination Experience -0.010
(1.289)
Interaction African American -0.198
(0.599)
Interaction Ethnic Minorities 0.356
(0.571)
Befriended African American 0.701
(1.108)
45
Befriended Other Ethnicity 0.242
(1.179)
Likes Ethnic Celebrities 0.691
(1.178)
Likes Ethnic Music 0.843
(0.921)
Likes Ethnic Food 0.635
(1.120)
Follows a Religion 0.312
(1.319)
Spirituality 0.028
(0.313)
Employed*Info -0.308
(1.900)
Employed*Reflect -1.226
(1.890)
Employed*Relate 0.809
(1.936)
Age*Info 0.732
(0.547)
Age*Reflect 1.094**
(0.549)
Age*Relate 0.698
(0.560)
Male*Info 0.417
(1.772)
Male*Reflect -0.622
(1.827)
Male*Relate 0.576
(1.840)
Married*Info -0.370
(1.238)
Married*Reflect -0.105
(1.319)
Married*Relate 1.385
(1.348)
Republican*Info 0.001
(2.224)
Republican*Reflect -3.301
(2.227)
Republican*Relate -1.449
(2.261)
Democrat*Info -3.085
(2.104)
46
Democrat*Reflect -3.229
(2.177)
Democrat*Relate -1.811
(2.175)
Discrimination Experience*Info 3.132*
(1.696)
Discrimination Experience*Reflect 2.799
(1.717)
Discrimination Experience*Relate 0.542
(1.774)
Perceived Chance of Receiving
0.054***
Money
(0.010)
Constant -4.474
(3.318)

Observations 1,013
Note: Tobit model was applied to estimate the stated donation with multiple interaction terms.
The robust standard errors are in parentheses. *, **, and *** denote that the coefficient is
significant at 0.1, 0.05, and 0.01 level, respectively.

47

You might also like