Abstract
People’s compassion responses often weaken with repeated exposure to suffering, a phenomenon known as compassion fatigue. Why is it so difficult to continue feeling compassion in response to others’ suffering? We propose that people’s limited-compassion mindsets—beliefs about compassion as a limited resource and a fatiguing experience—can result in a self-fulfilling prophecy that reinforces compassion fatigue. Across four studies of adults sampled from university students and online participant pools in the United States, we show that there is variability in people’s compassion mindsets, that these mindsets can be changed with convincing information, and that limited-compassion mindsets predict lower feelings of compassion, lower-quality social support, and more fatigue. This contributes to our understanding of factors that underlie compassion fatigue and supports the broader idea that people’s beliefs about the nature of emotions affect how emotions are experienced. Together, this research contributes to developing a strategy for increasing people’s capacity to feel compassion and their social support.
Many of us think that compassion drains us, but I promise you it is something that truly enlivens us.
In their book Compassionomics, Stephen Trzeciak and Anthony Mazzarelli (2019) argued that humanity faces a “compassion crisis” and that our collective lack of compassion is the world’s biggest problem. Compounding this compassion crisis is evidence suggesting limits to people’s compassion capacities after repeated exposure to suffering (Figley, 1995; Najjar et al., 2009). Although this view depicts compassion as a depletable resource, the limited and fatiguing nature of compassion may also be governed by mindsets (Job et al., 2010) that compassion is limited and fatiguing (i.e., a limited mindset), leading people to experience compassion as such. If so, then changing compassion mindsets represents an opportunity to overcome compassion’s limits. Thus, the present research tests whether mindsets about compassion as limited or nonlimited affect compassion and compassion fatigue.
Compassion and Its (Potential) Limits
Compassion is the feeling of concern for others’ suffering and the accompanying motivation to help (Goetz et al., 2010). Like other psychological capacities (e.g., short-term memory; Cowan, 2016), there may be limits to people’s compassion capacities. Relevant to the present research, compassion can weaken in response to prolonged, repeated exposure to suffering (i.e., compassion fatigue; Figley, 1995). Compassion fatigue was conceptualized as an occupational hazard among healthcare professionals, and it is defined as “a healthcare practitioner’s diminished capacity to care as a consequence of repeated exposure to the suffering of patients, and from the knowledge of their patient’s traumatic experiences” (Cavanagh et al., 2020, p. 640) and “a state of exhaustion . . . as a result of prolonged exposure to compassion stress” (Figley, 1995, p. 253).
Although compassion fatigue often stems from actively caring for patients, the mere knowledge of or exposure to patients’ suffering can also cause compassion fatigue (Cavanagh et al., 2020). Thus, media scholars also theorized about compassion fatigue following people’s repeated exposure to news media depicting the suffering of distant others disconnected from the self (Kinnick et al., 1996; Moeller, 1999). Supporting this idea, one experimental, lab-based study showed that participants who saw many (vs. few) compassion-inducing videos exhibited less empathy and reduced intent to help people in need during an unrelated task (Süssenbach, 2018). Although this context is different from health care, research in both contexts demonstrates the core elements of compassion fatigue: repeated exposure to suffering reduces subsequent compassion.
Why Mindsets May Influence Compassion
Although the capacity to feel sustained compassion in response to ongoing suffering may be limited, it is also possible that believing compassion is limited (i.e., having a limited mindset) contributes to compassion fatigue. In their review of limited willpower mindsets, Bernecker and Job (2019) explain that people with limited mindsets view their willpower as a “limited resource that gets depleted whenever used,” whereas those with nonlimited mindsets “reject this view and rather believe that using their willpower can even activate their mental stamina.” In the present research, people with limited mindsets may believe that feeling compassion depletes their emotional resources, requiring rest and recovery; people with nonlimited mindsets may disagree, and potentially view an initial experience of compassion as emotionally energizing, increasing their ability to feel compassion for others.
Critically, research suggests that limited mindsets can produce mindset-consistent experiences: Job and colleagues (2010) showed variability in limited versus nonlimited mindsets, the malleability of mindsets in response to information, and effects of limited mindsets on self-control and mental fatigue. Subsequent research suggests that these effects may be due to mindsets influencing people’s task-specific expectations and interpretations of mental experiences when exerting willpower (Chow et al., 2015; Francis & Job, 2018). Similarly, we propose that limited mindsets for compassion increases compassion fatigue via increased expectations of compassion fatigue in contexts eliciting ongoing compassion.
Statement of Relevance
Compassion has health and well-being benefits for the self and others. Unfortunately, people sometimes experience compassion fatigue—a decreased ability to feel compassion—when they are repeatedly exposed to people suffering. Thus, the present research explores a factor that can mitigate compassion fatigue: changing people’s compassion mindsets. Our research suggests that when people believe compassion is fatiguing and a limited resource, they experience more compassion fatigue and provide lower-quality social support; however, when people believe compassion is energizing and not limited, they feel less compassion fatigue and provide higher-quality social support. We also show that people can change their limited-compassion mindsets and become less susceptible to compassion fatigue. Altogether, this research cautions people against assuming they will experience compassion fatigue and to allow for the possibility that compassion for someone in need can be an energizing experience that motivates people to care about others in need, too.
Research Overview
We first established that people’s compassion mindsets vary, that limited-compassion mindsets predict compassion fatigue in a lab-based task, and that task-specific compassion fatigue expectations mediate this relationship (Study 1). We then probed this relationship in a new sample experiencing ecologically valid compassion fatigue because of COVID-19: We tested the stability of compassion mindsets across time, whether mindsets predicted compassion fatigue months later, and whether compassion mindsets buffered against compassion-fatigue risk factors (Study 2). Finally, we tested the malleability of compassion mindsets by having participants listen to a podcast that described compassion as either a limited or nonlimited resource and then testing whether experimentally-induced compassion mindsets affected compassion fatigue (Study 3) and social support (Study 4). For all studies, we obtained university Institutional Review Board approval and complied with our country’s legal guidelines for human-subjects research.
Study 1
Method
Participants
Participants from the United States on Prolific’s online pool received $1.50 to participate in a study about their beliefs. The design and analysis plan were preregistered (see the Supplemental Material). We recruited 320 participants. We initially aimed to have 80% power to detect a small-to-medium correlation (r) of .20 (requiring 193 participants), which is in line with the effect sizes of other mindsets on predicted outcomes (e.g., Burnette et al., 2020; Crum et al., 2013). We oversampled this number to account for potential exclusions. Thirty-two additional participants signed up despite not finishing the survey. Of the 352 participants, 42 were excluded on the basis of preregistered exclusion criteria (self-reported effort and attention), leaving us with a final sample of 310 participants (54.5% female; 80.6% White, 2.6% Asian, 3.2% Black, 10.6% Latino, 1.9% multiracial, 1.0% other; Mage = 26.1 years), 95% confidence interval (CI) = [25.2, 27.1]), and 94% power to detect a correlation (r) of .20.
Procedure and materials
After providing informed consent, participants rated their agreement with four items on a scale ranging from 1 (strongly disagree) to 6 (strongly agree). The items used to measure their compassion mindsets were adapted from Job et al. (2010): “Feeling compassion for others exhausts your resources, which you need to refuel afterwards” (reverse-scored); “After feeling sincere compassion for others, your emotional energy is depleted” (reverse-scored); “Feeling compassion is emotionally energizing, and afterward you are able to immediately start feeling compassion toward other people, too”; and “Even after feeling deep compassion, you can continue to feel compassion towards others.” These items were averaged to create a composite measure of compassion mindset (Cronbach’s α = .73).
After the compassion-mindset scale, participants answered demographic questions and a free-response question and scales about moral beliefs and decisions (collected for the purpose of piloting materials for an unrelated project; see the Supplemental Material for the full questionnaire). We included these other items and the demographics questions following the compassion-mindset scale to disguise the relationship between the compassion-mindset scale and the compassion-fatigue task. Next, participants began the second portion of the survey, again provided informed consent, and were introduced to the compassion-fatigue task. We explained to participants that the task was assisting a global public-interest organization and that it would involve seeing a series of nine images depicting “people experiencing difficult situations, such as illness, war, or displacement or animals that are suffering or in need” and answering questions about those images.
Before the task, we asked participants about their anticipated compassion fatigue specific to the upcoming task (“To what extent do you anticipate that feeling compassion for all of the people will feel . . .”). Participants reported these expectations using eight 6-point bipolar scales which we averaged to create a single score of compassion fatigue; higher scores indicated greater fatigue (M = 3.28, 95% CI = [3.18, 3.38], Cronbach’s α = .86). The eight scales were negative ↔ positive, exhausting ↔ energizing, depressing ↔ uplifting, difficult ↔ easy, like something you wanted to avoid feeling ↔ like something you wanted to feel, costly ↔ beneficial, like something you were unable to fully do ↔ like something you were able to fully do, and distressing ↔ calming. An exploratory factor analysis using a minimum residuals extraction and an oblimin rotation indicated that these eight items loaded onto a single factor.
Participants then continued to the task that was designed to induce compassion fatigue. On each image’s page, participants read a caption and answered, “To what extent do you, personally, feel an emotional response of compassion when looking at this photo?” Participants responded used a slider bar ranging from 0 (none at all) to 100 (an extreme amount). Following this task, participants reported the compassion fatigue they experienced using the same eight bipolar scales used to measure expected compassion fatigue (M = 3.95, 95% CI = [3.84, 4.05], α = .84). Finally, participants reported their attention and effort during the task.
Results
Did people vary in their beliefs about compassion as limited or nonlimited?
Beliefs varied among participants, with 29.7% reporting relatively limited mindsets (i.e., below the scale midpoint), 59.7% reporting relatively nonlimited mindsets (i.e., above the scale midpoint), and 10.7% at exactly the scale midpoint. On average, participants did not endorse a limited-compassion mindset (M = 3.95, 95% CI = [3.84, 4.06]). These responses were significantly higher than the scale midpoint of 3.50, t(309) = 8.31, p < .001, d = 0.47. Mindsets followed a normal distribution centered around the mean (see Fig. 1 for a histogram of Study 1 compassion mindsets; see Figs. S1, S2, and S3 in the Supplemental Material for histograms of compassion mindsets for Studies 2–4).

Histogram of participants’ limited-compassion mindsets in Study 1. On average, participants had endorsed nonlimited mindsets (sample mean of 3.95 was significantly higher than the scale midpoint of 3.5). Bin width is 0.5.
Are people’s beliefs about compassion associated with compassion fatigue?
As predicted, limited-compassion mindsets were correlated with greater experienced compassion fatigue (r = −.27, p < .001) and lower average compassion in response to the images (r = .29, p < .001), supporting the hypothesis that limited-compassion mindsets are associated with compassion fatigue.
In an exploratory analysis, we examined whether limited-compassion mindsets predicted decreasing compassion responses over time (i.e., image by image). To test this, we converted our data set to a long format and ran a linear mixed-effects model. We modeled a random intercept for each participant and fixed effects for compassion mindset, time (i.e., whether the image they saw was the first, second, third, etc.), and the Mindset × Time interaction, with this interaction indicating whether limited-compassion mindsets predicted decreasing compassion responses over time. We standardized predictor and outcome variables before entering them into the model. Compassion decreased more over time (i.e., image by image) for those with a relatively limited-compassion mindset, bmindset×time = 0.03, 95% CI = [0.01, 0.05], t(2,478) = 2.53, p = .01, consistent with the idea that limited-compassion mindsets predict greater compassion fatigue (see Fig. 2a).

Participants’ average compassion over time. Compassion in response to images of people suffering decreased for those with relatively limited-compassion mindsets, but not for those with relatively nonlimited-compassion mindsets (orange line). At left (a) we display results from Study 1; at right (b) we display results from Study 2. Shaded regions represent ± 1 SE.
Does expected compassion fatigue mediate the effect of compassion mindsets on compassion fatigue?
Consistent with predictions, two mediation models showed that expected compassion fatigue mediated the relationship between compassion mindsets and experienced compassion fatigue and average compassion in response to the images (Fig. 3).

Mediation models showing that expected compassion fatigue during the task mediates the relationship between compassion mindsets and two outcomes: experienced compassion fatigue and average compassion. Statistics for the average-compassion mediation model are shown in parentheses.
Study 2
In Study 2, we tested the relationship between limited-compassion mindsets and compassion fatigue in a different sample facing real demands on their compassion because of the COVID-19 pandemic. Study 2 also examined the consistency of mindsets across time, the relationship between limited-compassion mindsets and compassion-fatigue outcomes months later (i.e., a more conservative test of the effects of limited-compassion mindsets), and whether compassion mindsets buffered against compassion-fatigue risk factors.
Method
Participants
Students recruited from a large midwestern university received $10 per survey to participate in a multiwave study about their experiences and health during COVID-19. The analysis plan was preregistered (see the Supplemental Material for anonymized copy). A total of 2,164 people participated in the study, although only 1,162 people answered all the questions for our analyses from the five time points (99.9% power to detect a correlation of r = −.27, the value of correlation between compassion mindsets and self-reported fatigue in Study 1; 70.8% female; 58.1% White, 3.8% Black, 30.8% Asian, 0.2% American Indian, 5.6% mixed race). The high rate of dropout was likely due to three related factors: (a) the survey length (the baseline survey contained 192 items and took about 25 min; the monthly surveys had 127 items and took about 18 min; the exit survey had 60 items and took about 5 min); (b) the fact that there were five surveys spread across four months; and (c) our decision to exclude participants who did not complete measures at all five time points, which offered many chances for exclusion (results did not change if we did not exclude participants on the basis of this criterion). Men (50.7%) and participants marking “other” for gender (72.2%) were excluded slightly more often than women (43.9%), F(2, 2159) = 6.76, p = .001, η p 2 = .006, but there were no differences across race, F(4, 2101) = 0.66, p = .62, η p 2 = .001, or school year, F(6, 2156) = 1.76, p = .10, η p 2 = .005. There were also no differences between participants included and excluded for any of the variables used in our analyses (baseline compassion mindsets; all ps > .06 and all ds < .11).
The study had several other purposes not relevant to the present research (e.g., activity tracking), and thus eligibility for study participation included several criteria: age 18 years or older, being a confirmed undergraduate or graduate student, being able to provide digital informed consent, being comfortable with reading and speaking English, and having access to necessary resources for participating in an mHealth technology-based intervention (i.e., smartphone or tablet device and Internet access) while also being willing to use personal equipment or the Internet for the study. The sample size was determined on the basis of hypotheses from the data set unconnected to the present research.
Procedure and materials
The study protocol has been previously published with more details (Cislo et al., 2021). Participants were recruited from September to December 2020. Some students were recruited through flyers in campus buildings, but most were recruited through emails from the school’s registrar’s office. All study activities were conducted remotely with no in-person contact, and all study materials were mailed to participants’ residences. Interested participants who contacted the study team by telephone or email received additional study information. Following confirmation of university student status, the research coordinator emailed interested students an informed consent to electronically sign. Data for the present research was collected using an app with an integrated Qualtrics survey. Participants completed a baseline survey (Time 0); three monthly surveys (Times 1–3), with the first survey coming 1 month after baseline; and an exit survey around 2 weeks following the final monthly survey (Time 4). In the present research, we used compassion mindsets measured at baseline and averaged across the time points (Time 0–Time 3), compassion-fatigue outcomes, and risk factors measured in the exit survey. See the Supplemental Material for the full baseline, monthly, and exit surveys. Our analysis plan were preregistered (see the Supplemental Material for anonymized copy; a link to an identifiable copy will be provided on article acceptance).
Compassion mindsets
The compassion-mindset measure was the same measure as Study 1, except that the scale points ranged from 1 (not at all) to 5 (always). This difference between the Study 1 and Study 2 scales was a survey-design error, but the items still displayed good reliability at baseline, Cronbach’s α = .83, M = 3.44, 95% CI = [3.39, 3.49]. In addition, there was good reliability across the four time points that compassion mindsets were measured, Cronbach’s α = .91, M = 3.42, 95% CI = [3.37, 3.47], supporting the idea that mindsets are relatively stable.
Compassion fatigue
Participants answered the five items on a scale ranging from 1 (strongly disagree) to 7 (strongly agree): “Over the course of the COVID-19 pandemic, my compassion response has weakened over time”; “It has become harder to truly care about the COVID-19 pandemic as more and more people have become affected”; “I feel emotionally exhausted as a result of the COVID-19 pandemic”; “During the COVID-19 pandemic, I have felt more and more compassion as I am exposed to more people suffering” (reverse-scored); “Over time, my motivation to address people’s suffering from COVID-19 has increased” (reverse-scored).
In our preregistered analysis plan, we had decided that if the Cronbach’s α was under .70, we would run an exploratory factor analysis (using minimum residuals extraction, oblimin rotation, and factors based on parallel analysis) and retain all items that loaded onto the first factor to make a composite. The reliability of the five-item scale was a Cronbach’s α of .66. The first factor retained the two reverse-coded items, and reliability of the scale created from these two items was a Cronbach’s α of .70, M = 3.63, 95% CI = [3.56, 3.69] (the preregistered scale). However, because those two items described only the opposite of compassion fatigue, we also created a four-item composite from our scale that dropped the third item from the above list (Cronbach’s α = .72, M = 3.67, 95% CI = [3.60, 3.73])—the exploratory scale. This exploratory scale provided the benefits of a scale that (a) included items mapping onto compassion fatigue and (b) has a Cronbach’s α greater than .70.
Care provider risk factor
Participants responded to the question, “To what extent have you provided care or support for people suffering from COVID-19’s negative effects (e.g., illness, loss of a loved one, job loss)?” on a scale ranging from 1 (I have not done this at all) to 5 (I do this nearly every day or more), M = 2.27, 95% CI = [2.21, 2.33].
Distressing-media risk factor
Participants answered the question, “How frequently do you engage with news media that talks about the suffering and hardship that COVID-19 has caused?” on a scale ranging from 1 (less than once a month) to 7 (more than once a day), M = 4.51, 95% CI = [4.42, 4.61].
Results
Do compassion mindsets and compassion fatigue risk factors predict compassion fatigue?
Limited-compassion mindsets, whether measured at baseline or averaged across the four time points, were correlated with greater compassion fatigue for both the preregistered scale (rbaseline = −.31, rall = −.41) and the exploratory scale (rbaseline = −.23, rall = −.30), all ps < .001.
The compassion-fatigue risk factors had counterintuitive relationships with compassion fatigue. More time spent caring for those suffering from COVID-19 was associated with less compassion fatigue for both the preregistered (r = −.13) and exploratory (r = −.14) scales; likewise, more time spent engaging with news media depicting suffering from COVID-19 was associated with less compassion fatigue for both the preregistered (r = −.12) and exploratory (r = −.17) scales, all ps < .001.
Do nonlimited-compassion mindsets buffer against the effects of compassion-fatigue risk factors?
We preregistered tests of whether compassion mindsets moderate the relationship between each risk factor and compassion fatigue. However, because our risk factors were associated with less compassion fatigue, the theoretical model of nonlimited-compassion mindsets buffering against the influence of compassion-fatigue risk factors on increased compassion fatigue does not make sense (because the risk factors we conceptualized as risk factors did not function as risk factors). Thus, we present the results of the buffering-hypothesis analyses in the Supplemental Material (see Tables S1–S8) and summarize here that there was no evidence that compassion mindsets moderated the relationship between the two compassion-fatigue risk factors and compassion fatigue.
Discussion
Studies 1 and 2 revealed variability in compassion mindsets. Participants reported slightly more nonlimited (vs. limited) mindsets, on average; most mindsets were slightly above or below the scale midpoint; and mindsets followed a normal distribution.
Additionally, limited-compassion mindsets predicted more compassion fatigue in a lab-based task and an ecologically valid strain on compassion. Study 1 showed that these effects were mediated by increased situation-specific expectations of compassion fatigue. Study 2 also demonstrated that compassion mindsets are consistent across time and predict compassion fatigue months later (note that we did not preregister analyses of mindset reliability across time). Nonlimited-compassion mindsets did not buffer against compassion-fatigue risk factors in Study 2 (e.g., providing care to someone suffering, engaging with distressing news media), but these risk factors were negatively correlated with compassion fatigue, making it difficult to examine this buffering hypothesis. Because the findings from Studies 1 and 2 were correlational, we designed an experiment in Study 3 to test compassion mindsets’ malleability and their causal effect on compassion fatigue.
Study 3
Method
Participants
Participants from the United States on Prolific’s online pool received $2.00 to participate in a study that required listening to an audio clip. The design and analysis plan were preregistered (see the Supplemental Material for anonymized copy). Because of our power analysis, we planned to have at least 200 participants to achieve 80% power to detect a medium effect size (d = 0.40), but we recruited 272 participants because we anticipated exclusions. Twenty-seven participants signed up despite not finishing the survey. Of the 299 participants, 53 were excluded on the basis of preregistered exclusion criteria (i.e., attention-relevant manipulation checks and self-reported effort and attention), leaving us with a final sample of 246 participants (50.8% female; 74.0% White, 11.8% Asian, 5.7% Black, 5.3% Latino, 1.6% multiracial, 0.4% American Indian, 1.2% other; Mage = 37.2 years, 95% CI = [35.6, 38.8], 80% power to detect Cohen’s d = 0.36).
Procedure and materials
After providing informed consent and completing an audio check to make sure sound was working on their computers, participants completed Part 1 of the study. For this, they were randomly assigned to listen to an 8.5-min podcast about compassion as limited versus nonlimited, which had inspired previous mindset research that had manipulated people’s beliefs using convincing articles (Chiu et al., 1997; Lee et al., 2019; Schumann et al., 2014) and film clips (Crum et al., 2013). We created the podcast for this study, which we called “Fast Feels: The Podcast That Explains How Emotions Shape Our Lives in 10 Minutes or Less.”
Both the limited and nonlimited conditions used the same set of three voices—the host and two guests. In both versions, the host set up the episode about the limited/nonlimited nature of compassion during the COVID-19 crisis. The host then interviewed two actors, with brief host monologues in between. The first actor talked about her reaction to the various pandemic-related tragedies, such as job losses, death counts, and her coworkers’ difficulties taking care of their children. The second actor was a doctor who talked about his reaction to treating patients and how it has affected his life at home. The experiences recounted in the podcast were ecologically valid—they were generated from our qualitative analysis of people’s self-reported compassion-fatigue experiences from separate pilot studies that we ran before and during the global pandemic. Below we provide some examples of the manipulation (full scripts are available in the Supplemental Material). For example, in the limited condition, the host says: . . . There are limits to people’s capacity to feel compassion . . . It means that people cannot truly feel compassion for all of COVID’s victims. It means that people helping those suffering from COVID-19 day in and day out will ultimately feel emotional fatigue. Ultimately, people will run out of compassion. It’s human nature.
In the nonlimited condition, however, the host says: . . . People can feel unlimited amounts of compassion. What does that mean—unlimited amounts of compassion? It means that people can feel compassion for all of COVID’s victims, even when they haven’t met them. It means that people can help those suffering from COVID-19 day in and day out without any feelings of emotional fatigue. Ultimately, compassion can fuel itself. It’s human nature.
Later, one of the guests (a doctor) explains in the limited condition: My compassion for my patients . . . has carried me through this . . . I’m a more compassionate doctor with my last patient of the day than with my first—it’s as if my compassion has a life of its own, growing stronger with each patient I see.
In the unlimited condition, the doctor says: It’s extremely exhausting to feel compassion for my patients . . . I’m a more compassionate doctor with my first patient of the day than with my last—it’s as if I have a bank account of compassion, and every time I see a patient, I’m withdrawing from my compassion account.
After the podcast, participants answered two manipulation-relevant attention checks and some questions about their attitudes toward the podcast (see the Supplemental Material for all questions) before being redirected to what was ostensibly a second, separate survey. Participants provided informed consent again and completed the same image-viewing task from Study 1. Following the task, participants reported their compassion fatigue using the same scale as in Study 1 (α = .87). Finally, they were asked about their attention and effort and their demographics.
Results
Are people’s compassion mindsets malleable?
Consistent with other work on mindsets, our experimental manipulation was able to change participants’ compassion mindsets consistent with the information they received. Participants had significantly more limited-compassion mindsets in the limited condition, M = 3.05, 95% CI = [2.89, 3.21], versus the nonlimited condition, M = 3.78, 95% CI = [3.60, 3.95], t(237.9) = 6.04, p < .001, d = 0.77. Compassion mindsets among those in the limited condition were significantly below the scale midpoint, t(130) = 5.56, p < .001, d = 0.49; compassion mindsets among those in the nonlimited condition were significantly above the scale midpoint, t(114) = 3.10, p = .002, d = 0.29.
Do limited-compassion mindsets (vs. nonlimited-compassion mindsets) reduce compassion and increase compassion fatigue?
Consistent with predictions and conceptually replicating Study 1, participants in the limited (vs. nonlimited) condition felt less average compassion and more compassion fatigue. Participants’ average compassion in response to the images of suffering was significantly lower in the limited condition, M = 75.3, 95% CI = [71.8, 78.9], versus the nonlimited condition, M = 81.2, 95% CI = [77.6, 84.7], t(242.9) = 2.29, p = .02, d = 0.29, suggesting that a limited-compassion mindset was associated with lower levels of compassion. In addition, participants experienced more compassion fatigue in the limited condition, M = 4.02, 95% CI = [3.85, 4.19], versus the nonlimited condition, M = 3.62, 95% CI = [3.41, 3.82], t(228.5) = 2.98, p = .003, d = 0.38. Interestingly, participants in both conditions reported compassion-fatigue scores above the scale midpoint—limited condition: t(129) = 6.08, p < .001, d = 0.53; nonlimited condition: t(114) = 3.10, p = .002, d = 0.29.
As an exploratory analysis, we examined compassion responses over time (i.e., image by image) using two linear mixed-effects models. The first model used the same inputs as the model in Study 1, and results replicated those of Study 1: Compassion decreased more over time for those with a relatively limited-compassion mindset—bmindset×time = 0.03, 95% CI = [0.01, 0.06], t(1,966) = 2.88, p = .004 (see Fig. 2b). The second model replaced “mindset” with “condition” as the fixed factor. Here, the extent to which compassion decreased over time for those in the limited (vs. nonlimited) condition did not differ significantly—b = 0.01, 95% CI = [−0.01, 0.03], t(1,966) = 0.96, p = .34—suggesting that participants in the limited (vs. nonlimited) condition did not have significantly greater decreases in image-by-image compassion. This null finding can likely be attributed to the fact that measured mindsets (vs. experimental condition) are a more proximate and stronger predictor of compassion-fatigue outcomes. In other words, there will always be some variability in people’s mindsets within a condition (i.e., those who maintain a nonlimited mindset in the limited condition), making “condition” a noisier and less precise operationalization of compassion mindsets and thus less capable of detecting smaller effects. Supporting this idea, the effect size capturing the relationship between measured mindsets and self-reported compassion fatigue and lower average compassion is larger than the effect size of condition on those two outcomes.
Discussion
Study 3 suggests that compassion mindsets are malleable. Moreover, participants in the limited (vs. nonlimited) condition felt less compassion and more self-reported compassion fatigue in response to the pictures of people in need. Limited-compassion mindsets (as measured by our scale) predicted decreases in compassion over time, but participants in the limited (vs. nonlimited) condition did not show greater decreases in compassion over time. Studies 1, 2, and 3 did not explore behavioral effects; we address this in Study 4.
Study 4
Method
Participants
Participants from the United States on Prolific’s online pool received $4.00 to participate in a study about “COVID experiences.” We needed 146 participants to achieve 80% power to detect the Study 3 effect size of condition on experienced compassion fatigue (d = 0.38). We oversampled this target to account for potential attrition and the fact that other dependent variables may have smaller effect sizes; we recruited 286 participants. Of these, 34 failed a manipulation-relevant attention check, leaving a final sample size of 252 (47.2% female; 61.6% White, 10.0% Asian, 10.8% Black, 6.8% Latino, 0.4% Native Hawaiian or Pacific Islander, 9.2% mixed race, 1.2% other; Mage = 32.1 years, 95% CI = [30.6, 33.6]; 80% power to detect Cohen’s d = 0.35). This study was not preregistered.
Procedure and materials
After providing informed consent and completing an audio check to make sure sound was working on their computers, participants were randomly assigned to listen to the limited or nonlimited podcasts from Study 3. Next, they answered two manipulation-relevant attention checks and some questions about their attitudes toward the podcast before being instructed to proceed to the next portion of the survey. Participants then completed a filler task (three questions from the Comprehensive Assessment of Rational Thinking; Stanovich, 2016) before moving on to the task designed to measure social support.
For the helping task, we asked participants to give feedback on job materials to some of our students whose job opportunities had been affected by COVID-19. We explained to participants that many students had lost their jobs and internships, including those attending the universities affiliated with the research team. Participants then read the following: Many of our students are sharing their experiences online as a way of networking and setting themselves up for other job offers. Typically, in their posts they are explaining how their job offer got rescinded, highlighting their skills, and sharing their resumes . . . With the permission of our students, we will be sharing drafts of these kinds of posts that our students will be posting to their LinkedIn accounts . . . You will have the opportunity to give feedback on the drafts and offer words of encouragement to the student . . . You will have the chance to help with up to 6 students. After each student, you can choose to go to the next student or to advance to the demographics portion of the survey.
For each student, participants answered three questions capturing their social support for the student. First, participants had the opportunity to offer emotional and informational social support by giving free response answers to these two prompts: “Below, you can offer your sympathies and write some words of encouragement for the student” (emotional social support) and “Below, you can give suggestions for ways for the student to improve their post. Again, this can be about anything or on any part of their materials” (informational social support). On average, participants wrote 22.7 words of emotional support (SD = 14.9) and 17.5 words of informational support (SD = 13.9) for the first student they helped.
Two coders rated the quality of emotional and informational social support on a scale ranging from 1 (terrible) to 5 (excellent; Cronbach’s α = .74 and α = .83, respectively). Their responses were averaged for data analyses. Complete coding directions are available in the Supplemental Material.
Next, participants indicated their intentions to offer instrumental social support by answering the question, “If you saw this post on your LinkedIn feed, how likely would you be to try and connect the student to someone in your network?” on a scale from 1 (definitely would not) to 5 (definitely would). After these three questions, participants indicated whether they would like to “help the next student” or “skip ahead to the next portion of the survey.”
As soon as they chose “skip ahead,” or after helping six students, participants completed the compassion-mindset scale (the same as in the previous studies; α = .85), questions about COVID-19, and questions about demographics. The compassion-mindset scale in this study contained two additional items (i.e., six items total), but to be consistent across studies, we computed compassion mindsets using only the four items that were present for the previous studies; including the two additional items did not affect results.
Results
Are people’s compassion mindsets malleable?
Participants had significantly more limited-compassion mindsets in the limited condition, M = 2.83, 95% CI = [2.60, 3.06], versus the nonlimited condition, M = 3.88, 95% CI = [3.69, 4.06], t(192.2) = 7.10, p < .001, d = 0.92.
Do compassion mindsets affect social support?
We analyzed the effect of our manipulation on social support using several methods. First, we computed the separate averages for each type of social support (emotional, informational, and instrumental) for each participant, only including ratings for students whom participants chose to help (as opposed to entering “0” scores for students who did not receive help). We found that participants offered significantly less social support in the limited (vs. nonlimited) condition. See Table 1.
Social Support Across Conditions in Study 3
Note: CI = confidence interval.
Next, we tested whether participants in the limited (vs. nonlimited) condition helped more students. Contrary to our predictions, participants in the limited condition offered emotional or informational support to numbers of students, Memotional = 1.78, 95% CI = [1.45, 2.11], Minformational = 1.67, 95% CI = [1.33, 1.83], similar to those in the nonlimited condition, Memotional = 1.89, 95% CI = [1.62, 2.16], t(198.6) = 0.50, p = .61, d = 0.07, Minformational = 1.58, 95% CI = [1.33, 1.83], t(189.5) = 0.42, p = .68, d = 0.05. This suggests that there were no effects of podcast condition on the total number of students helped. This could have been a floor effect, given that most participants helped only one student. Because participant payment was not tied to extra time spent helping the students, it is possible that the benefits of finishing the study more quickly outweighed motivations to help for participants in all conditions (mean completion time for the whole study was 28.5 min, with a standard deviation of 13.9 min).
We had also planned to examine whether participants’ quality of social support changed over time, but there were insufficient numbers of participants who helped with multiple students to meaningfully analyze this (19.0% and 17.5% of participants provided emotional and informational support to three or more students, respectively).
Discussion
Study 4 suggests that changing compassion mindsets can affect behavior: Participants in the limited (vs. nonlimited) condition offered lower-quality emotional and informational social support and gave less hypothetical instrumental social support to students in need of career help. Contrary to predictions, the mindset manipulation did not affect the number of people participants helped. This finding could be explained by the fact that nonlimited-compassion beliefs may not increase people’s perceptions of nonpsychological resources (e.g., time), which was required for participants helping more people.
General Discussion
Do mindsets about compassion’s limited or nonlimited nature affect people’s experience of compassion fatigue? Across four studies, we found that beliefs about compassion as a limited or nonlimited resource vary across people, are malleable in response to convincing information, and predict compassion and compassion fatigue in response to COVID-19 and lab-based tasks.
Theoretical contributions
Prior research has identified individual differences that predict compassion response, including trait compassion (Hou et al., 2017), attachment style (Mikulincer et al., 2005), and social class (Stellar et al., 2012). The compassion-fatigue literature has also identified antecedents of compassion fatigue, including gender (Sprang et al., 2007), work setting (Sprang et al., 2007), and traumatic experiences (Hinderer et al., 2014). Our research adds to this list by suggesting that people’s preexisting beliefs about the limited or nonlimited nature of compassion can also influence compassion and compassion fatigue. This is powerful, because the malleability of mindsets (compared with other individual differences) suggests an avenue for increasing compassion and reducing compassion fatigue.
Our findings also add to a growing body of work connecting mindsets to emotions and fatigue. Prior research connected mindsets to stress (Crum et al., 2013; Huebschmann & Sheets, 2020), self-compassion (Chwyl et al., 2021), and willpower (Job et al., 2010). One article particularly relevant to the present investigation demonstrates that people’s belief that empathy can be developed predicts effortful responses when empathy is challenging (Schumann et al., 2014). Our work departs from this work in several ways. First, we focus on compassion (vs. empathy). Although some have suggested that empathy is involved in the elicitation of compassion (Goetz et al., 2010; Hou et al., 2017), empathy differs from compassion because it involves the vicarious experience of another’s emotions and has different neural signatures (Klimecki et al., 2014). Second, Schumann et al.’s work focuses on the belief that empathic ability can be developed through practice, whereas the present research focuses on beliefs about compassion as a limited resource that is fatigued with use.
Although our research suggests that mindsets influence compassion and compassion fatigue, it does not reveal whether compassion has objective limits or suggest there are no limits to other resources (e.g., time, money). Depending on how one operationalizes compassion’s limits, there are likely limits to compassion capacities. For instance, research on the different, but related, phenomenon of the collapse of compassion shows that people typically feel less compassion as the number of people suffering increases (Cameron & Payne, 2011); it seems implausible that any intervention would lead compassion to linearly increase with the amount of suffering people perceive. Thus, more research is needed to understand compassion’s true limits.
Implications for science communication
Our research suggests that science media may induce self-fulfilling prophecies. Many high-profile news articles have reported compassion’s limits (Carey, 2011; Fraga & Crowe, 2020; Petrow, 2019; Resnick, 2017), and this may contribute to people’s beliefs about, and experiences of, compassion. It is important for scientists and science journalists to consider the potential negative impacts of their work (Turnwald et al., 2019) and communicate the nuanced and uncertain nature of scientific findings (Lewis & Wai, 2021).
Limitations
Our research had several limitations. First, in our correlational studies, participants may have interpreted the scale items as statements about their personal experiences of compassion, rather than as “generic you” statements (Orvell et al., 2017) reflecting general mindsets; if participants interpreted the items as the former, then our results may reflect individual differences in people’s experiences of compassion rather than mindset effects. However, the effect of the experimental manipulation (which talked about the general nature of compassion) on people’s scale responses suggests that people interpreted the scale as asking about general compassion beliefs. Future research on mindsets should distinguish and measure the relationships between objective individual differences, self-perceptions, and general beliefs about some capacity (e.g., compassion’s limits). Although general mindsets inform self-perceptions (Chow et al., 2015), it remains unclear whether people’s self-perceptions of their experiences inform general mindsets and whether these self-perceptions stem from genuine individual differences. Our scale was also limited in that we adapted a preexisting scale (which exports that scale’s benefits and drawbacks); future researchers could rigorously develop a new compassion-mindset scale.
Second, demand characteristics and social desirability may have influenced participant responses. For instance, social desirability may have been a third variable underlying the relationship between nonlimited-compassion mindsets and higher compassion responses. Likewise, participants may have been motivated to give condition-consistent responses in the experimental studies, given that the postmanipulation tasks occurred shortly after the podcasts. With that said, several features of our studies reduced demand effects. First, we observed effects of mindsets on compassion and compassion fatigue in studies without experimental manipulations. which suggests that relationships between mindsets and compassion outcomes are not solely attributable to condition-specific demand effects. We also reduced demand by examining how compassion mindsets affected outcomes 4 months into the future (Study 2); using filler tasks to disconnect mindset measurement and manipulation from outcome measures (Studies 1 and 4); and ostensibly obtaining participants’ consent for inclusion in new studies (i.e., separate-tasks paradigm, Study 1 and Study 3; Kehner et al., 1993). Future research could further reduce demand and social desirability with more subtle mindset manipulations (e.g., biased scales; Job et al., 2010) and social-desirability control variables.
Third, we examined compassion fatigue in a different context from most compassion-fatigue research (i.e., health care). In Study 1 and Study 3, we examined compassion fatigue in response to repeated exposure to pictures of others’ suffering; in Study 2, we examined students’ compassion fatigue in response to COVID-19. This operationalization of compassion fatigue is similar to how compassion fatigue has been conceptualized by scholars studying repeated exposure to news media depicting suffering (Kinnick et al., 1996; Moeller, 1999) and to the lab study examining how repeated exposure to videos depicting suffering led to compassion fatigue (which they called “moral fatigue”; Süssenbach, 2018); however, it is qualitatively different from the health-care context, in which people are actively caring for those suffering. Future research should explore whether nonlimited-compassion mindsets buffer against compassion fatigue among health-care professionals.
Fourth, our measures of compassion fatigue could be improved. For instance, the self-reported compassion-fatigue scale includes items about “ability” and “motivation” to feel compassion. We included both because prior compassion-fatigue scales had included items related to motivation (e.g., Eng et al., 2021), and our results did not change when the motivational items were removed. However, it could be argued that the motivational (vs. ability) items are less connected to weakened compassion capacity and are thus beyond the scope of compassion fatigue.
Future directions
Future researchers could explore the origins of people’s compassion mindsets, such as exposure to media on the topic, mindset-consistent personal experiences, or social learning (for a review on the origins of lay theories, see Haslam, 2017). Qualitative research could also reveal these origins, as well as a richer understanding of people’s compassion mindsets. For instance, our research focused on beliefs about compassion depleting emotional resources and sustaining over time, but some people may have focused on the number of people they can feel compassion for simultaneously.
Future researchers could also consider other outcomes flowing from compassion mindsets, such as self-compassion, which has been linked to mental-health benefits (Neff, 2011) and prosociality (e.g., Neff & Pommier, 2013). It is unclear, for instance, whether people believe that other-focused compassion and self-compassion involve necessary trade-offs and whether this belief is influenced by people’s limited-compassion mindsets. Future research could also examine whether limited-compassion mindsets influence the collapse of compassion (Cameron & Payne, 2011) and related phenomena, such as the identifiable-victim effect (Kogut & Ritov, 2005), compassion fade (Butts et al., 2019), and psychic numbing (Slovic, 2007).
Finally, future research could explore compassion mindsets in different contexts. Our sample was generally WEIRD (Western, educated, industrialized, rich, and democratic; Henrich et al., 2010), American, and White (and disproportionately female in Study 2). Given research showing that Chinese (vs. American) students believe willpower to be less limited (Sun et al., 2019), compassion mindsets may differ across cultures. Another situational factor of note is whether the context involves people feeling compassion for those experiencing the same tragedies as themselves (e.g., many people during COVID) or those suffering a different harm (e.g., compassion for factory-farmed animals). This distinction modulates compassion response (Ruttan et al., 2015). Experiencing tragedy alongside others may elicit zero-sum thinking and activate a limited mindset, or perhaps feeling compassion for others experiencing different tragedies activates a limited mindset by bringing more suffering into focus. Our results supported our hypotheses both when participants were experiencing a shared tragedy as compassion targets (e.g., Study 2) and when they were not (e.g., Studies 1 and 3), but future work could more precisely examine the moderating role of this factor.
Conclusion
Viktor Frankl (1992, p. 75) famously wrote in 1946, “We who lived in concentration camps can remember the men who walked through the huts comforting others, giving away their last piece of bread. They may have been few in number, but they offer sufficient proof that everything can be taken from a man but one thing: the last of human freedoms—to choose one’s attitude in any given set of circumstances—to choose one’s own way.” Consistent with Frankl, our research suggests that with the right mindset, people can sustain their compassion amid repeated exposure to suffering. In a world with many people in need and many opportunities to help them, nonlimited-compassion mindsets may reduce compassion fatigue and foster one’s capacity to help.
Supplemental Material
sj-pdf-1-pss-10.1177_09567976231194537 – Supplemental material for Compassion Fatigue as a Self-Fulfilling Prophecy: Believing Compassion Is Limited Increases Fatigue and Decreases Compassion
Supplemental material, sj-pdf-1-pss-10.1177_09567976231194537 for Compassion Fatigue as a Self-Fulfilling Prophecy: Believing Compassion Is Limited Increases Fatigue and Decreases Compassion by Izzy Gainsburg and Julia Lee Cunningham in Psychological Science
Footnotes
Acknowledgements
We thank Molly Ketch, Lansing O’Leary, Ying Li, Linnan Zhou, and Sara Soroka for their help with this project.
Transparency
Action Editor: Paul Jose
Editor: Patricia J. Bauer
Author Contribution(s)
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
