Charity preferences and perceived impact moderate charitable giving and associated neural response

Abstract

Charitable giving depends on individuals’ abilities to make altruistic decisions. Previous studies suggest that altruism involves recruitment of neural resources in regions including social processing, reward/reinforcement learning, emotional response, and cognition. Despite evolutionary and social benefits to altruism, we know that humans do not always engage in altruistic behavior, like charitable giving. Understanding the underlying processes leading to decisions to donate is vital to improve prosocial community engagement. The present study examined how characteristics of the charitable giving opportunity influence an individual’s decision to give and the neural engagement underlying these features. Twenty-nine participants subjectively rated ten charities on their value, effectiveness, and the subject’s personal chance of donating. Participants then completed an fMRI task requiring them to decide to donate to certain charities given the probability of the donation helping, their personal preference for the charity, and whether the donation came at cost to themselves. There was a significant reduction in donating when the probability of helping was low versus high, and subjects were significantly less likely to donate to their lowest-rated charities. Further, probability of a donation being helpful and how much the subject favored a charity moderated PCC and left IFG engagement. Interestingly, reward neurocircuitry did not demonstrate similar sensitivity to these variations. These results may suggest individuals engage motivated reasoning to justify failure to donate, while donations are driven by emotion mentalizing that focuses on the welfare of others. This may provide valuable insight into how to engage individuals in altruistic giving.

1. Introduction

Charitable activities are a major part of social life across the world. According to the Urban Institute’s Nonprofit Sector Brief 2018, there are more than one million public charities registered in the United States with a total of $1.9 trillion in annual revenue (McKeever, 2018). Of this amount, private individuals gave $410 billion. The largest portion of these charities focus on human services (35%; e.g., homeless shelters, foodbanks), but also represent causes like education, the arts, health, and the environment. Individuals also contribute through volunteering; in 2017, 25% of Americans volunteered a typical 50 h per year each, accounting for $445 billion in volunteer time. Charitable activities address essential problems; for example, the United States Department of Housing and Urban Development reported that over 500,000 people are estimated to be homeless nightly, of which shelters help up to 65% (2019). Although there are some differences in patterns of charitable giving and volunteerism across the world, partially driven by economic development and policies (Einolf, 2017), many other countries have similar patterns of giving, volunteerism, and focus on health and human services (Turcotte, 2015).

What is it that makes individuals participate in charitable activities? Charitable giving is a prosocial behavior. Empirical data has long supported empathy or care for others as a source of motivation for charitable giving (Andreoni et al., 2017) and altruism in general (Batson and Shaw, 1991). For example, subjects with higher trait empathetic concern have been shown to be more likely to give up money to prevent another from pain (FeldmanHall et al., 2015) and general self-report of actual charitable giving (Kim and Kou, 2014). Indeed, experimental work shows that more salient requests for donations lead to increased volunteerism through increased feelings of empathy and social responsibility (Kandaurova and Lee, 2019). However, these studies also found that social guilt was a mediating factor, suggesting a more complicated set of mechanisms.

There is further reason to suspect that charitable giving is not exclusively driven by empathy. Antithetical to care-based altruism, and despite the clear evidence for willingness to give to charitable foundations, less than 2% of charitable donations are between individuals (McKeever, 2018). Real-world charitable giving is more complicated than even the most ecologically valid laboratory model, but even those reveal rates of giving are sensitive to a variety of factors. The presence of observers (Izuma et al., 2010), social pressure (DellaVigna et al., 2012), fewer overhead costs (Gneezy et al., 2014), identifiability of the victim (Small and Loewenstein, 2003), matching grants (Karlan and List, 2007), and guilt arousal (Hibbert et al., 2007) increase charitable giving. Charitable giving can also be influenced by the availability of incentive programs, like tax write-offs (List, 2011).

Accounts of the neural substrates of altruism may explain further these observed psychosocial factors. Tusche and colleagues suggest three neural processes contribute to charitable giving in distinct and dissociable ways: empathy for beneficiaries of charity corresponding to anterior insula engagement; taking the perspective of others corresponding to temporoparietal junction (TPJ) engagement; and general attention shifting localized to the superior temporal sulcus (STS; Tusche et al., 2016). Other neuroimaging research suggests a more complex interaction between neural circuits. For example, Hare and colleagues found a hub role for the medial prefrontal cortex (mPFC) in the willingness to donate to charity, finding that engagement was correlated with the value of the donations given while connectivity to STS and anterior insula positively associated with decisions to donate. Further, meta-analysis reveals engagement of additional regions in charitable giving, including the temporal pole, anterior and posterior cingulate (A/PCC), thalamus and striatum (VS) regions, possibly suggesting an interaction between social, emotional, and valuation processes during decisions about charitable giving (Cutler and Campbell-Meiklejohn, 2019). Finally, the amygdala, a key hub in coordination with the mPFC in integrating social, emotional, and cognitive information, may also play a role in charitable giving (Marsh, 2016).

Neural processes may also be moderated by characteristics of the giving scenarios. In the presence of observers, and when the recipient was depicted in a photo along with the request, charitable giving was associated with greater VS response, suggesting a role for the reward of explicit social approval in altruism (Genevsky et al., 2013; Izuma et al., 2010). Voluntary compared to mandatory prosocial giving was also associated with greater VS, as well as insula, response (Harbaugh et al., 2007). Helping ingroup versus outgroup individuals engaged the anterior insula modulated by trait empathetic concern (Hein et al., 2010). Social distance of recipients also modulated connectivity between TPJ and mPFC (Strombach et al., 2015).

There remain questions about how characteristics of charitable causes influence neural processing during decisions to donate or not. It is vital to understand what contributes to increased charitable giving in order to identify strategies to encourage altruistic behavior and begin to address outstanding social problems like homelessness and food insecurity. Neuroimaging can be particularly helpful in this field, where individuals may try to present themselves in the best light by misrepresenting (intentionally or implicitly) their motivations for giving.

In this study, we looked at the neural correlates of charitable giving, particularly examining how subjective value of a charitable cause, probability of the donation helping, amount and source of donation influence decisions to donate and neural engagement during deliberation about decisions to give to charity. We hypothesized that (i) charity preference and (ii) a high probability of helping the cause would lead to (a) higher donation rates and (b) greater engagement of social and reward regions, particularly when (iii) the cost to self was little to none. Moreover, we predicted (iv) that more engagement of these circuits would be associated with decisions to donate as opposed to not donating.

2. Methods

2.1. Participants

Participants (n = 29) were individuals enrolled in studies through the National Institute on Alcohol Abuse and Alcoholism. Half of participants (n = 15) were in the inpatient treatment program for alcohol use disorder (AUD). This sample was selected to provide pilot data on potential substance use related differences in prosocial processing, though group differences were not the primary aim of this study. We have included a sensitivity analysis in the Supplemental Materials to verify the main effects examined here do not differ qualitatively between groups. All participants completed NIAAA’s screening protocol, including providing basic demographic information, and were determined eligible. Exclusion criteria included left handedness, neuromotor difficulties, claustrophobia, MRI contraindications, use of psychotropic medications, positive urine drug screens, current symptoms of alcohol withdrawal and pregnancy (if applicable). See Table 1 for sample descriptives, including associations between demographic variables.

Table 1.

Demographic characteristics.

A) Characteristics of Sample
	descriptives
	mean		sd
Age (Years)	42.07		10.26
Income Level	4.79		3.03
Education (Years)	14.83		3.27
Gender	%Male:		57.14
AUD status	%AUD:		51.72
B) Associations Between Demographics
	correlation		t
	age	income level	gender	AUD
Age (Years)	-	-	− 0.18	0.24
Income Level	0.15	−	0.55	− 3.05*
Education (Years)	0.26	0.61***	− 0.63	− 2.74*

Notes: A) descriptive statistics for demographic characteristics; B) correlations between age, education, and income level. Association between income levels and other demographics tested with spearman’s rho; other correlations are pearson’s r. Significance indicated as follows:

^*p < 0.05,

^***p < 0.001.

2.2. Procedures

All procedures were carried out in accordance with protocols approved by the NIH Addictions Institutional Review Board. After providing written, informed consent and a day-of urine drug/pregnancy screen, participants read descriptions of 10 charitable causes. The included charities were well established foundations selected to represent a variety of topics (i.e., medical care, hunger, housing, the environment, children’s issues) and affiliations (i.e., international/American, religiosity). Each description was approximately 150–250 words taken from the marketing materials/websites of each organization and was paired with an icon of the charity’s logo. The last charitable cause was described as “Direct Donation to a Person in Need”, and its explanation included a description of the ways in which a direct donation to a person soliciting money on the street (i.e., a “panhandler”) might help, a description of who a typical “panhandler” is, and why he or she might solicit money. The icon representing this cause was a picture of a cardboard sign stating “Homeless, Please Help, Thank You”.

After the subjects read the descriptions of each charity, they were asked to complete a Charitable Foundations Questionnaire (CFQ). This questionnaire asked subjects to rate each charity on “How valuable is the cause the charity supports?“, “How effective is the charity at making a difference towards helping the cause?“, and “What is the chance that you would use your own money to donate to the charity?“. Responses were on a scale of 1–9, where 1 was very poor and 9 was excellent. Subjects were also asked to rank the charities overall from their most to least favorite.

Imaging Task:

After completing the CFQ, participants completed two runs of a Charitable Giving Task during an fMRI scan. For each trial, subjects were presented with a giving choice scenario and asked to indicate if they wanted to donate. The did so by pressing a button to indicate “Accept” or “Reject” during the 6 s for with the scenario was displayed. After making the choice, subjects were given feedback of whether their donation “helped” or “failed to help” (feedback displayed for 3 s), randomized based on the actual probability displayed on the choice screen. Responses of “Reject” are always followed by feedback of “failed to help”. Between trials within the blocks, a fixation cross was displayed for a random interval between 0.5 and 2 s. Each run consisted of 45 trials and lasted between 9 and 10 min. See Fig. 1 for a diagram of the task timeline and choice elements, also described below.

Fig. 1. Diagram of Altruism Task.

Task structure including A) timing, C) block types, charity features, and D) icons used to represent each charity included. B) Example choice presentation as seen by participants is also shown.

Choices were presented in three block types based on scenario type: Giving-Self (where donation came from personal account at cost to self), Giving-Foundation (where donation came from an organizational account at no cost to self), and Getting (where money was added to the participant’s personal account). The Getting scenario was a monetary gain condition designed to control for general monetary decision making; Giving scenarios included some choices with no potential gain or loss, operationally defined as Neutral for this analysis. Each run consisted of three blocks of each type (9 total blocks) and block order was randomized within run.

Within blocks, choices varied on the following elements: Amount of donation ($1 or $10), Probability of helping (30%, 90%, or 100%), and charity Preference (favorite, middle, and least favorite, as ranked on the CFQ). The charitable cause receiving the potential donation was cued by the icon used in the charity description; subjects were given the opportunity to again review the charity descriptions immediately before entering the scanner and told to make sure they were familiar with the icons. These choice elements were randomized within block, and not all combinations of choice elements were present in each block. See Supplemental Materials for instructions read to participants that elaborated on each of these elements.

Imaging Data Acquisition:

Participants were scanned on a Siemens 3 T Prisma Magnetic Resonance Imaging (MRI) machine at the NIH Clinical Center in Bethesda, MD. Imaging data were acquired during the Charitable Giving Task using a standard echoplanar-imaging pulse sequence (TR: 2000 msec, TE: 30 msec, flip angle: 90°, FOV: 24 × 24 cm, 38 mm slice thickness, 36 slices, multi-slice mode: interleaved). We also collected a T1-MPRAGE structural scan for the purposes of coregistration (TR: 1900 msec, TE: 3.09 msec, flip angle: 10°, FOV: 24 × 24 cm, 1 mm slice thickness, 144 slices, muli-slice mode: single shot).

Image Processing:

The fMRI data for each participant was first processed on a single subject basis using AFNI (v16.2.16; Cox, 1996: afni.nimh.nih.gov). Time courses were shifted for each voxel to be aligned to the same temporal origin. Outliers were identified in the time-series at this point. Volumes across the time series were aligned to the base volume and to the skull-stripped anatomy of the subject; we then used non-linear warping to transform the data to standard MNI space. Finally, we used a 4 mm full-width at half-maximum Gaussian kernel to smooth the data and scaled the resulting data to a run mean of 100 with values falling between 0 and 200. Outlying TRs and those with a motion derivative value at or above 0.3 mm were censored from future analyses. Blur estimates were calculated based on the average correlation across masked brain; this autocorrelation function (ACF) was used to calculate cluster size significance thresholds as described in the Group Level Analysis section.

Stimuli onset times and their durations were regressed using a duration modulated block function to identify signal associated with each task condition across runs; demeaned motion parameters and their derivatives were also included to remove variance associated with movement. Given the small number of trials of each combination of task parameters, we decided a priori to model our task in four separate single-subject level models based on stimuli type: 1) Scenario type [Giving-Self; Giving-Foundation; Getting] * Choice behavior [Accepted; Rejected]; 2) Scenario type [Giving-Self; Giving-Foundation] * Probability [30%;90%;100%]; 3) Scenario type [Giving-Self; Giving-Foundation; Getting] * Amount [$0/$1;$10]; and 4); Preference [favorite; middle; least favorite]. For Model 1 only, we also modeled Feedback type [“Successfully helped”, “Failed to help”, “Got $0”, “Got $10]. These stimuli definitions did not vary across models, so were not duplicated.

Group Level Imaging Analysis:

Whole-brain group-level imaging results were analyzed using multivariate modeling (3dMVM) in AFNI. Task condition (corresponding to single-subject models described above) was included as a within-subject variable. Separate group-level analyses were run for each of the 4 model types described in the preceding paragraph. All group-level analyses included age, income level, years of education, sex, and AUD status as between-subject covariates. Analyses were restricted to a group mask, created using the intersection of the MNI 152 template and the individual subject masks.

Our imaging hypotheses (coded b in the introduction) were examined statistically using the following seven (7) contrasts of interest, modeled within the overall models. (Model 1; hypothesis iii & iv) 1-Giving [Self + Foundation] > Neutral Feedback; 2-Giving [Self + Foundation] > Neutral Scenario; 3-Giving [Self + Foundation; Accepted] > Giving [Self + Foundation; Rejected]; 4-Giving [Self] > Giving [Foundation]. (Model 2; hypothesis ii) 5-Giving [Self + Foundation; 100%] > Giving [Self + Foundation; 30%]; (Model 3; hypothesis iii) 6-Giving [Self + Foundation; $10] > Giving [Self + Foundation; $1]; and (Model 4; hypothesis i) 7-Giving [Favorite] > Giving [Least Favorite]. Neutral Scenario was operationally defined as Getting scenarios when $0 was offered; Neutral Feedback was operationally defined as the feedback display following such trials of “Got $0”.

For the purposes of our study, contrast 1 and contrast 2 corresponded to general charitable or altruistic processing and were run to verify the ability of the task to elicit charitable or altruistic decision making processes as commonly understood in the field; contrast 3 was designed to address whether different neural circuits are engaged when individuals ultimately decide to donate to charity. Contrasts 4–7 allowed us to address our primary hypothesis, that “charity preference and a high probability of helping the cause would lead to higher donation rates and greater engagement of social and reward regions, particularly when the cost to self was little to none.”

Results for each model were thresholded at voxel-wise p < 0.001 and a cluster extent of k = 12 using third nearest neighbor clustering and two-sided tests, calculated via 3dClustSim to control for multiple comparisons (using a family-wise error [FWE] rate correction approach) at a level of alpha = 0.05. 3dClustSim determines these thresholds by using averages of each subject’s spatial ACF estimates to simulate noise.

Behavioral analyses and supplemental plots of imaging analyses were carried out using R/RStudio (v. 3.4.2). ggplot2 (v. 2.2.1; Wickham, 2009) in R was used to plot behavioral and supplemental imaging results.

Behavioral Analysis:

Behavioral data was parsed by first identifying the frequency of “Accept” choices for each scenario for each subject during the imaging task; this (i.e., frequency of decisions to donate) was the primary behavioral dependent variable. From the CFQ, we identified average charity valuation, sense of charity efficacy, and likelihood of making monetary charitable donations by averaging these ratings across charities for each subject.

Effect of task conditions on frequency of decisions to donate was tested in three separate ANCOVA models using the rstatix package in R (Kassambara, 2020). The within subjects variable differed by model to address the hypotheses (coded a in the introduction) as follows: 1) (hypothesis i) Preference [favorite; middle; least favorite]; 2) (hypothesis ii) Probability [30%;90%;100%]; and 3) (hypothesis iii) Scenario Type [Giving-Self; Giving-Foundation]. These models included age, income level, sex, AUD status, and years of education as covariates. Continuous covariates were mean centered. Degrees of freedom were corrected, where sphericity assumptions were not met in the within subjects factor, using Greenhouse-Geisser correction. When more than two levels of significant within subjects variables were present, posthoc tests were conducted using pairwise t-tests, adjusted for multiple comparisons using Bonferroni correction.

In order to explore the potential influence of the specific charity characteristics and subjective impressions of the charitable causes, we conducted several exploratory analyses of the CFQ data. Each of the 10 charitable cause was explored in terms of subject preference; χ² tests were used to examine the frequency with which each was chosen by the subjects as least, middle, and most favorite, and correlations were calculated to examine the association between preference rank and CFQ variables. We also examined the correlations between frequencies of decisions to donate to charity and CFQ variables.

3. Results

3.1. Behavioral results

Participants responded with a button press to 95.93% of choices (sd: 7.35%; i.e., less than two trials per run), suggesting adequate understanding of and attention to the task. Participants decided to donate to a charitable cause in 69.32% of scenarios (sd: 12.77%; see Table 2 for a summary of behavioral responses and responses on the CFQ). There was a significant effect of probability of donation helping the cause (F (1.15,26.54) = 50.98, η²_G = 0.598, p < 0.001) and individual preference for the charity on likelihood of donation (F (1.3,30) = 21.96, η²_G = 0.375, p < 0.001; see Fig. 1). Posthoc tests revealed differences between all probability classes (certain versus high: difference = 4.59%, p_adj = 0.042; low versus certain: difference = 50.75%, p_adj < 0.001; low versus high: difference = 46.15%, p_adj < 0.001) and between all preference classes (favorite versus middle: difference = 8.38%, p_adj = 0.006; least favorite versus favorite: 32.07%, p_adj < 0.001; versus middle: 23.68%, p_adj = 0.002). There was not a significant effect of source of money on frequency of decisions to donate to charity (F (1,23) = 3.509, _η²_G = 0.010, p = 0.074).

Table 2.

Charitable giving behaviors.

	Mean	SD
Frequency of Decisions to Donate to:
Charity, Overall	69.32	12.77
Charities with 30% Chance of Helping	36.62	29.36
Charities with 90% Chance of Helping	84.14	15.63
Charities with 100% Chance of Helping	88.47	13.47
Favorite Charity	83.25	15.05
5th Favorite Charity	74.49	13.50
Least Favorite Charity	51.10	29.05
Charity from Personal Bank	68.36	13.03
Charity from Foundation	70.68	13.04
Frequency of Decisions to Take from Charity	42.14	25.58

Notes: Means and standard deviations of the frequency of “Accept” button presses on the behavioral task of charitable giving.

There was no significant effect of CFQ variables on rates of decisions to donate during the imaging task. Charity preference rank was highly correlated with CFQ variables and CFQ variables were also highly correlated with each other (See Table 3). There was also a significant difference in the frequency with which each charity was selected as the favorite, middle or least favorite (χ² (18) = 78.567, p < 0.001). Participants most frequently selected St. Jude Children’s Research Hospital as their first choice (p_adj p < 0.001, freq = 15), while they most frequently choose direct donation to a person in need as their least favorite option (p_adj = 0.008, freq = 13). Feeding America and Habitat for Humanity were more likely to be ranked as #5 (middle favorite; Feeding America: p_adj = 0.003, freq = 7; Habitat for Humanity: p_adj = 0.027, freq = 8).

Table 3.

Charitable foundations questionnaire responses.

	Value of Charity	Effectiveness of Charity	Chance of Donating
A) Descriptives:
Mean	7.59	7.03	6.25
SD	1.19	1.33	1.68
B) Correlations:
Effectiveness (rho)	0.775*	-	-
Chance of Donating (rho)	0.644*	0.760*	−
Donation Choice Freq (rho)	− 0.034	− 0.079	0.117
Preference Rank (tau)	0.508*	0.542*	0.592*

Notes: A) Means and standard deviations of the average responses on the Charitable Foundations Questionnaire (CFQ). Responses were on a scale of 1–10 (with 1 being the lowest and 10 being the highest). B) Spearman (rho) correlations between CFQ variables and with frequency of button press choices to donate to charity, and Kendall (tau) correlations with ranking of charity preference. Significance at p < 0.001 indicated with

^*No other correlations were significant at a level of p < 0.05.

3.2. Imaging results

There was a main effect of charitable giving scenarios (see Table 4/Fig. 2). In contrast 1 (Giving [Self + Foundation] > Neutral Feedback), there was extensive engagement of regions across the brain including right fusiform gyrus, left supramarginal gyrus, frontal eye fields, bilateral thalamus, posterior cingulate (PCC), caudate/putamen, bilateral anterior insula/IFG, and bilateral hippocampus. There was greater deactivation in the superior temporal sulcus (STS; x = 1, y = −43, z = 25, T = 4.716, k = 12). The PCC was also found in contrast 2 (Giving [Self + Foundation] > Neutral Scenario). Decisions to donate to charity compared to choosing not to donate to charity was associated with greater engagement of right IFG, left visual association area (V3), right supplementary motor area (SMA), and right superior parietal lobule (SPL) regions (see Table 3/Fig. 3).

Table 4.

Neural engagment in charitable giving.

Charitable Giving Scenarios > FB of Expected No Gain
Region	Ref	BA	x	y	z	T	k
Fusiform	1	37	28	−68	−11	11.61	1836
Primary Sensory/Supramarginal Gyrus	2	1/40	−53	−32	58	10.31	1238
Frontal Eye Field	3	8	4	25	43	10.17	274
Thalamus	4	*	−14	−20	10	8.36	191
Posterior Cingulate	5	23	1	−29	25	7.28	109
Caudate	6	*	13	10	4	9.02	89
Thalamus	7	*	10	−14	10	5.4	29
Inferior Frontal Gyrus/Insula	8	45/13	−32	19	10	5.89	29
Superior Temporal	9	22	52	−17	−8	−6.72	26
Gyrus/Superior
Temporal Sulcus
Hippocampus	10	*	−23	−29	−5	7.44	24
Hippocampus	11	*	25	−23	−5	5.79	20
Insula	12	13	40	19	−5	5.46	20
−	13	*	−23	−32	25	−5.21	15
Charitable Giving Scenarios > Expected No Gain Scenarios
Region		BA	x	y	z	T	k
Posterior Cingulate		23	1	−44	25	4.72	12
Charitable Giving scenarios: Accepted > Rejected
Region		BA	x	y	z	T	k
Inferior Frontal Gyrus (pars opercularis)		44	40	13	31	6.03	29
Visual Area 3		19	−20	−89	−5	5.16	23
Precentral Gyrus		6	31	−2	52	5.4	15
Superior Parietal Lobule		7	34	−50	46	5.79	12

Notes: Coordinates represent cluster peaks, and are reported in MNI space. BA-Brodmann area. Ref - Reference number corresponds to label on Fig. 2 images of results.

Fig. 2. Frequency of Decisions to Donate by Charity Features.

Error bars represent standard error. * indicates significant differences, as reported statistically in the text.

Fig. 3. Neural Processing during Charity Scenarios > Feedback of Expected No Gain.

(left) Images of whole-brain results, thresholded at p < 0.001, k = 12. Warm colors indicate a positive T value, while cool colors indicate a negative T value. X,Y,Z values represent position on slices shown. Numbers 1–13 in squares correspond to Table 4, referencing the specific cluster labels and statistics. (right) Y axis indicates beta values averaged across significant clusters as plotted in the left graph. Error bars represent standard error. Top plot betas correspond to activations averaged across all clusters with significant positive T values. Bottom plot betas correspond to activations averaged across all clusters with significant negative T values. Patterns for averages as plotted in this graph are representative of patterns when individual significant clusters are plotted. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

There were also effects of charity characteristics on neural underpinnings of charitable giving. Contrast 5 (Giving [Self + Foundation; 100%] > Giving [Self + Foundation; 30%]) showed engagement of the PCC (x = 16, y = −62, z = 22, T = 5.784, k = 15; see Fig. 4A). Contrast 7 (Giving [Favorite] > Giving [Least Favorite]) was associated with greater engagement of left IFG and reduced deactivation of the PCC (see Table 5/Fig. 4B). There were no significant findings in contrast 4 (Giving [Self] > Giving [Foundation]) or contrast 6 (Giving [Self + Foundation; $10] > Giving [Self + Foundation; $1]) (Fig. 5).

Fig. 4. Neural Processing during Charity Scenarios, Accepted > Reject.

X and z values on brain images represent slice coordinates at the plotted cluster. Warm colors indicate a positive T value. Brain images shown thresholded at p < 0.001, k = 12. Letters in squares correspond to bar plots on the right. On bar plots, error bars represent standard error. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

Table 5.

Neural engagment moderated by charitable scenario features.

Probability of Helping: Certain > Low
Region	BA	x	y	z	T	k
Posterior Cingulate	31	16	−62	22	5.78	15
Charity Preference: Favorite > Least Favorite
Region	BA	x	y	z	T	k
Posterior Cingulate	23	−11	−59	22	−6.02	21
Inferior Frontal Gyrus (pars orbitalis)	45	−50	22	−11	−4.83	16

Notes: Coordinates represent cluster peaks, and are reported in MNI space. BA-Brodmann area.

Fig. 5. Neural Processing during Charity Scenarios by Scenario Features.

X and z values on brain images represent slice coordinates at the peak of plotted clusters. Warm colors indicate a positive T value, while cool colors indicate a negative T value. Brain images shown thresholded at p < 0.001, k = 12. On bar plots, error bars represent standard error. Other scenario features not plotted in A or B were not significant. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

4. Discussion

Individual altruistic behavior can play an important role in addressing the suffering of humanity. We know individuals can be induced to make prosocial donations to help causes in need of support through social pressure and other incentives, but understanding the neuropsychological drives of charitable giving may allow us to improve engagement of individuals in prosocial behaviors. To address this, we examined the neural correlates of charitable giving decisions in the absence of external motivators. We largely replicated previous charitable giving studies by finding clusters of activation including basal ganglia, middle frontal, temporal/insular, and parahippocampal regions. Moreover, we confirmed our hypotheses that donations to preferred charities and those with a high probability of helping the cause would be increased. Further, donating to a preferred charity and uncertainty that a donation would help was associated with decreased PCC engagement, as well as decreased IFG engagement in the latter condition. However, we did not find that reward processes were sensitive to these variations.

4.1. Influence of charity preference on charitable giving

Individuals were 32% less likely to agree to donate money to their least favorite charity compared to their most favorite. Given this difference was not as large in the comparison with a charity of middling preference, it is likely this is driven by a tendency to not want to donate to that charity, rather than a desire to donate to the preferred one. This decrease in donations corresponded to increased IFG engagement and decreased PCC deactivation when deciding whether to donate to their least favorite charity. This IFG region has been engaged consistently across altruistic compared to strategic giving decision studies (Cutler and Campbell-Meiklejohn, 2019). Generally, the left IFG is thought to be involved in selection of competing semantic knowledge (Thompson-Schill et al., 1997), so it follows that when participants are considering donating to a cause they don’t favor, but that has some merit, greater left IFG engagement may reflect increased internal debate over various semantic information; this uncertainty is reflected in the observed 50% donation rate for these charities. For charities where individuals have already endorsed the merits of the cause, there is no need for this selection and evaluation of information.

The PCC results are also consistent with this explanation, though suggests a more nuanced explanation; part of the default mode network (Buckner et al., 2008), task activation tends to be anticorrelated with PCC engagement, although evaluative activity does tend to engage this region (Vogt et al., 1992). The PCC’s internally-focused evaluative function suggests the possibility that individuals relied more on motivated reasoning when deciding whether to donate to causes they dislike. Motivated reasoning, a form of implicit emotion regulation when individuals are asked to reason about a topic that threatens a strong internalized opinion (in this case, whether to donate to a disliked charity), has been previously associated with greater engagement of the PCC (Westen et al., 2006). Motivated reasoning has also been shown to be used to justify not donating to charity (Exley, 2020). Here we found neural response associated with regions consistent with motivated reasoning only in cases where the cause was inconsistent with internalized preferences. Indeed, decisions not to donate in general were not associated with engagement of this PCC area; such a finding would have been inconsistent with an explanation of motivated reasoning.

At face, this is inconsistent with previous behavioral and economic work of “donor loyalty” (Sargeant and Woodliffe, 2007), as it seems driven by consistency in dislike rather than preference. Our behavioral results reflected no specific impact of favoritism on donations; there was no significant difference in chances of donating to the top and middle charities. On the other hand, the least favorite charity was associated with a significant decline in donations. However, an alternative interpretation focusing on the favorite charity as a special case might instead focus on the PCC deactivation as a correlate to more reasoning in general; previous work shows a lack of effect of cognitive load on altruistic decisions (Tinghög et al., 2016), perhaps suggesting donations to favorite charities are more automatic or habitual (i.e., “donor loyalty”, as cited above), and thus engage less altruistic processes. However, the behavioral results are more consistent with our first explanation in the preceding paragraph.

Importantly, this effect could also be driven by different types of charities being ranked by the participants as most and least favorite, rather than preference per se. Participants ranked St. Jude Children’s Research Hospital mostly highly of the options they were given, while they ranked Direct Donation to a Person in Need lowest. This pattern of responding means it is possible that the effect we are exploring here was not that of donating to a favorite compared to least favorite charity but instead, donating to an organization compared to an individual, to children compared to an individual, or to the cause of health compared to homelessness. The former two alternatives do not fit our data; half of individuals ranked organizational charities lowest and nearly half of top ranked charities were not those targeted towards children, so those characteristics do not drive the influence of preference on neural engagement.

The lattermost alternative cannot be ruled out given our dataset, though there was significant variability in ranking of causes; some individuals (n = 3) did rank “direct donation to a person in need” as their favorite cause. Though preference for medical charities is itself an interesting finding (Red Cross was also commonly ranked as a favorite charity, freq = 4), we hesitate to expand it as representative of a universal preference. Our subjects were, after all, volunteers for research in a medical setting. Previous research has indicated that individual values (and experiences) can influence preferences for certain charities (Bennett, 2003; Saxon-Harrold, 1993); greater PCC involvement in response to “direct donation to a person in need” could indicate reflection on personal experiences with encountering homeless persons asking for money, or familiarity with poverty (Reed, 1998). This explanation seems unlikely given our data; there was no association between socioeconomic status and charity preference, and by the same token, individuals would have direct experience donating to medical charities and in the case of half our subjects, receiving free medical care through medical research.

4.2. Influence of donation efficacy on charitable giving

A more ventral area of the PCC was engaged in charitable giving scenarios with a certain probability of helping the cause, compared to those with a low chance. This also corresponded to a 51% difference in likelihood of donating, with the low chance of helping scenarios leading to donations in only 38% of cases. Engagement of the ventral PCC has been associated with care ethics, i.e., identifying a person’s needs and aiming to address them via empathy and altruism (Robertson et al., 2007). PCC deactivation in low probability cases could suggest a sense of being unable to help a cause interferes with one’s engagement of care processes, leading to decisions not to donate. However, engagement of this region of the PCC has been implicated in a variety of processes, particularly self and other emotion mentalizing (Atique et al., 2011), representation of self versus others (Feng et al., 2018), and general internal focus as a hub of the default mode network (Buckner et al., 2008). These functions are not incompatible with the first explanation; previous studies have shown that compassion, a form of emotion mentalizing, increases altruistic behavior (Weng et al., 2013).

On the other hand, these latter PCC functions might imply the effect of low probability of helping was to interfere with emotional responses rather than increase altruism specifically; studies find that more emotionally salient solicitations to charity led to an increased likelihood of donation, mediated by individual experience of emotion and decreased deliberation (Small and Verrochi, 2009). The accuracy of the first versus second explanation could be important to real-world charitable giving because there is often a sense that individual donations won’t make a difference (e.g., $5 is nothing to a million dollar charity with administrative overhead, or giving change to an individual on a street corner won’t help them acquire housing); though our behavioral results suggests improving subjective perception of charity efficacy in solicitations might increase donations, this latter interpretation would imply general emotional appeals are more effective than appealing to reasoning (i.e., by identifying specific ways in which donations will help). Further research designed to tease apart this conundrum is necessary before any concrete conclusion is made.

It is possible that “certainty” rather than chances of helping, per se, drove this finding; however, we do not believe this to be the case. Participants were similarly likely to donate to a cause if there was a 90% chance of helping (i.e., high, but not certain), and when the donation came at no cost to the participant, there was no difference in PCC engagement between the certain and high probability conditions. Moreover, previous studies investigating the PCC and ambivalence/certainty indicated a ventral/dorsal correspondence, respectively (Luttrell et al., 2016); our certain condition engaged ventral PCC, further emphasizing the unlikeliness of this interpretation of our findings. We also do not believe this PCC finding or the corresponding behavioral effects were driven by risk. Participants could have endorsed donating in all scenarios if they had chosen; they were not told to conserve their funds and both the bank and foundation funds were larger than the possible amount to spend on donations. Therefore, there was no risk associated with accepting these low probability charity scenarios. Moreover, studies implicating PCC in processing of risk often find increased engagement during processing of risky scenarios (McCoy and Platt, 2005), again in contradiction to our findings.

4.3. Processes leading to decisions to donate versus not donate

There are also valuable insights into charitable giving that can be gleaned from the comparison of decisions to donate versus those to reject. Decisions to donate to charity compared to choosing not to donate to charity were associated with greater engagement of right IFG, supplementary motor (SMA), and superior parietal lobule (SPL) regions. These regions are part of the ventral attention network (Fox et al., 2006), playing an important role in attention switching and selection. Engagement of these regions during decisions to donate could reflect attention or perspective switching leading to deciding to give; switching from focus on personal goals to the needs of others has previously been forwarded as a key mechanism of altruism (Hare et al., 2010). Our results are consistent with that theory, and thus would suggests strategies for soliciting donations focusing on encouraging attention focus and perspective taking.

However, it could be that IFG, SMA, and SPL regions are not involved in the decision to give to charity per se, but rather charitable giving scenarios that ultimately lead to decisions to give engage those regions. For example, increased IFG here might mean increased retrieval of memories during donation scenarios that are ultimately accepted; previous work shows disruption of the right IFG leads to impaired memory control (Stramaccia et al., 2017). This would be consistent with work reporting individuals often donate to charities based on personal experiences with an issue (Bennett, 2003), but would likely not indicate that external moderation of IFG could change charitable giving behavior. On the other hand, training in compassion can increase SPL engagement during altruism (Weng et al., 2013), possibly suggesting a greater elicitation of compassion during the scenarios that are ultimately accepted. However, the SMA and SPL (particularly, the intraparietal sulcus) are also commonly implicated in motor/task sequence imagery (Jubault et al., 2007; Lacourse et al., 2005); their engagement in “accepted” donation trials may simply reflect the decision to pursue an action, rather than anything specific about charitable giving.

4.4. Influence of cost of donation on charitable giving

Unexpectedly, we did not observe hypothesized effects of size or source of donations on either frequency of decisions to donate or neural response during decision making. In other words, our participants were not sensitive to cost of donation. We also failed to observe modulation of value computation networks by charity features, despite previous studies consistently finding VS engagement differences associated with task elements like degree of social pressure (Genevsky et al., 2013; Izuma et al., 2010). This could be a byproduct of our task design: our scenarios were hypothetical and dealing with relatively small amounts of money ($1 or $10). As such, an explanation for this finding could be that our study failed to capture the value computation elements in general. Previous studies find hypothetical giving and other hypothetical decision making to produce similar patterns of responses (Johnson and Bickel, 2002; Nilsson et al., 2016), and participants in our study were instructed to make the decisions as if they were real, though it is possible that subjects did not do so. If this were the case, there would be no difference in terms of “cost” between a donation coming from their personal account versus a foundation, or the $9 difference between amounts.

Notably, our task in general did engage the expected neural responses that have been interpreted in the literature as reward/value computation. The main effects of considering whether to donate to a charity were consistent with those clusters identified in the most recent meta-analysis, including the bilateral thalamus, IFG, striatum, and hippocampus (Cutler and Campbell-Meiklejohn, 2019). Thalamic and striatal regions in particular are consistent with value computation being engaged during charitable processing.

Another explanation, then, is that variations in charitable giving, absent environment influences as in our laboratory study, are primarily driven by non-economic processes; there was no neural or behavioral correlates of economic features like amount or cost-to-self. As such, we found no evidence to support “warm-glow” altruism and similar cost-benefit motivations of charitable giving that might contradict the role of emotion and care in altruism. This is consistent with accounts of “extreme altruism” (helping others at potential extreme cost), finding these decisions are intuitive rather than a product of analysis (Rand and Epstein, 2014). Laboratory studies found real-world extreme altruists had a greater empathetic neural response than controls, also suggesting that care, rather than discounting of the cost of altruism, drives real-world altruism (O’Connell et al., 2019). Indeed, theory suggests that differences in perception of social closeness may drive willingness to behave altruistically (Vekaria et al., 2017).

4.5. Limitations

The first limitation of the present study was its power. Our sample size was relatively small (n = 29) and some contrasts were uneven or had low numbers of trials (e.g., there were less “Reject” trials than “Accept” for most subjects). Even some planned comparisons (e.g., Favorite > Least Favorite Charity) had a minimal number of trials per condition (20 each). Though this can decrease power in noisy data, the range of 20–30 trials per condition given our n is sufficient (Desmond and Glover, 2002; Huettel and McCarthy, 2001). This is not an easily addressable problem in design; Huettel and McCarthy estimate more than 100 trials per condition are needed to detect the majority of voxels activated by a task manipulation. We chose to use a smaller number of trials to minimize attention drift and fatigue in our subjects. The trade-off is a higher false negative rate, and less power to explore individual differences or more granular features in charitable giving situations (e.g., donations to organizations versus individuals in need). Importantly, we also did not investigate the association between CFQ variables and neural data because these were highly correlated with each other and with preference; including them in the models would have introduced a problem of multicollinearity.

Individual differences, which we attempted to control for statistically, likely have important influences on these processes. For example, some studies have shown individuals with lower socioeconomic status tend to be more charitable and generous (Piff et al., 2010), though paradoxically, others indicate that donors tend to be of higher education, in a marital relationship, and of a better subjective financial situation (Bekkers and Wiepking, 2011). Our sample includes 15 individuals diagnosed with and receiving treatment for alcohol use disorder (AUD). Although not yet established how AUD impacts charitable giving, if at all, and lacking the power to explore that in this sample, we are nevertheless confident that the within-subject effects seen here are consistent across our typically functioning and AUD samples. Sensitivity analysis showed similar patterns of hemodynamic response in the reported contrasts when examined in each group separately.

We were also limited to our charitable giving scenarios being hypothetical. This could mean that subjects did not process the giving scenarios the way they would in real world donation scenarios. However, the observed average frequency of decisions to donate to charity in the fMRI task was identical to participants average self-report of the chance they would use their own money to donate to each charity (6.25 on a scale of 9, or 69%). Although both hypothetical, the consistency between the behavioral and self-report measurements suggest this limitation didn’t significantly affect the results (FeldmanHall et al., 2012). Previous behavioral studies also find that intentions to donate, self-reporting donations, and actual donations are consistent (Nilsson et al., 2016). Moreover, although a previous study found neural differences between hypothetical and real altruistic behaviors, none of them overlap with the regions observed here; we cannot say for certain if charitable scenario features (probability of helping, favor towards charity, amount and source of donation) would have modulated neural engagement of other brain regions not observed here during real charitable giving.

We conducted this study to provide insights into the decision-making mechanisms in charitable giving. Indeed, we have identified several characteristics of charity scenarios that lead to reduced giving and moderation of neural engagement. However, we are unable to establish the directionality of these relationships; although we would expect the scenario characteristics to influence behavior via processes in the brain, it is possible the two are affected separately and distinctly. The only way to establish these principles causally would be to conduct a brain stimulation study, using TMS to knock out PCC or IFG regions, for example, and observing how this procedure affects behavior.

We are further limited in our understanding of mechanisms because we did not collect self-report data on factors influencing participants’ personal decisions to accept or reject the donation scenarios in the task. This means our attempts to discuss the neuroimaging results in terms of mental processes are at their core reverse inference (Hutzler, 2014; Poldrack, 2011). We attempted to minimize the issues with using reverse inference in the following ways: first, by considering multiple functions of the given regions and what these different functions might imply; second, ensuring that these functions were representative of the literature rather than the authors’ specific knowledge by using Neurosynth (Yarkoni et al., 2011), an automated meta-analytic tool (neurosynth. org); and third, discussing the consistency of the inferred function with behavioral and theoretical literature on charitable giving.

We examined neural activity during charitable giving scenarios in contrast to two different neutral conditions. We did this because we believed they had complimentary advantages and disadvantages. The first was neural activity time locked to presentation of feedback of gaining $0 on trials where they were presented with a $0 Getting scenario (i.e., expected no-gain). This was the most neutral condition that effectively controlled for baseline global activity and simple effects of reading words on the screen. This means that the contrast can still include all elements of a charitable giving, including decision making; however, this condition was modeled over 3 rather than 6 s, doesn’t require subjects to prepare for an action, and doesn’t present a visual icon, possibly allowing for signal of non-interest associated with these things to be captured by the contrast. Our second neutral was neural activity time locked to the presentation of that $0 Getting scenario. This contrast allows us to see the unique variance of altruism. However, this neutral condition includes decision-making and value computation, processes that are inherently part of deciding whether to give to charity. We were also concerned it would induce a short-lived negative reaction since subjects were being told they could not earn money, and thus influence the overall contrast. Overall, by including results from contrasts with both Neutrals, interpretation limitations can be mitigated somewhat.

5. Conclusions

We found evidence that individuals are more likely to donate to charity when they are confident in its efficacy and believe in the foundation’s work and cause. Our neuroimaging results suggest this is driven by social processing, particularly through focus of attention to consider the needs of others instead of personal benefits. The degree to which this neural switching occurs is affected by preferences for the charitable cause and the perception of its effectiveness; charitable causes where there was a certain chance of helping engaged more emotion mentalizing, while charities that were disliked engaged motivated reasoning to justify a lack of donations. Although hardships like homelessness, poverty, starvation, and violence continue to run rampant in the world, this support for care-based altruism is reason to be hopeful. As social creatures, humans have an innate motivation and neurobiological imperative to help those around them. We cannot hope to solve the world’s problems in a day but understanding what influences these altruistic decisions is a first step in engaging humanity for the public good.