The Power of Empathy: Experimental Evidence of the Impact of Perspective-Focused Interventions on Support for Prison Reform

Abstract

As a result of COVID-19, individuals have experienced situations that may help them relate to others, including more limited ability to interact with their environment. Thus, this survey experiment (N = 2,229) tests whether perspective-focused interventions can help increase support for prison reform. Findings suggest that perspective-getting (providing the perspective of an incarcerated individual via a narrative description of dealing with confinement) increased self-reported support for prison reform initiatives, compared with information only. In addition, a perspective-taking prompt—nudging participants to put themselves in the shoes of the incarcerated individual when reading their narrative—may help boost intention to take action in support of prison reform. Future avenues for research and implications are discussed.

Introduction

According to data from the Sentencing Project (2021), the United States incarcerates more of its population than any other nation in the world.¹ The “tough on crime” rhetoric and punitive justice system policies of the late-20th century—from Nixon’s “War on Drugs” to Reagan’s Anti-Drug Abuse Act—ushered in the era of mass incarceration, increasing the nation’s rate of confinement by more than 500% (The Sentencing Project, 2021). In response, prison reform efforts have often focused on tackling mass incarceration in the past few decades, such as California’s justice system realignment initiatives beginning in the last decade (Lofstrom & Martin, 2015). However, public support for these efforts is sometimes waning, and therefore, the success and continuation of reform efforts may be hindered by public dissention.

Finding effective methods for bolstering public support for these initiatives is crucial given mass incarceration’s negative impacts on our nation and the minimal incapacitation effect of incarceration on crime prevention (Bushway & Paternoster, 2009). One substantial effect of mass incarceration is swollen costs to the taxpayer—the estimated average cost of housing each incarcerated individual was approximately US$31,000 in 2010 (Henrichson & Delaney, 2012), and this realization has encouraged greater bipartisan support for efforts to reduce the number imprisoned or jailed (Bellamy et al., 2012). Another effect of over-incarceration—one often overlooked by the general public—is its harmful impact on public health outcomes. Although the literature has long-documented the negative effects of incarceration on health (Haney, 2003), the COVID-19 pandemic has more prominently exposed incarceration’s impact on exacerbating public health problems. The virus devastated U.S. prisons and jails. In 2020, about 20% of individuals in prison in the United States tested positive for COVID-19 and more than 1,700 died from the virus, leading corrections departments to release thousands of individuals from confinement (Associated Press, 2020). Like other reforms, support for these pandemic-initiated changes is vital to sustaining the proliferation and continued momentum of efforts to end the era of mass incarceration.

The COVID-19 pandemic and its mitigation efforts, including mask-wearing, social distancing, sheltering in place, and even self-isolating, have limited the level of control individuals have over their lives and their ability to interact with their environment. As a result of acquiring this newfound perspective, people may be more able to be empathetic toward individuals with vastly less control over how they interact with their environment—those who have been incarcerated during the outbreak. Therefore, it may be timely to consider perspective-focused interventions to garner support for prison reform efforts among the public. Perspective-focused interventions aim to help individuals understand the perspective of another and increase empathy for others, either through directly hearing others’ perspectives, considering how they may feel, or getting information about their points of view. Perspective-focused interventions have been applied in a variety of contexts, although to the author’s knowledge, no study has addressed whether these interventions can increase empathy for incarcerated individuals or increase support for improving the criminal justice system more broadly. To meet this empirical need, I conducted a survey experiment through Amazon’s Mechanical Turk platform to test whether various perspective-focused interventions can improve support for prison reform, as measured through self-reported endorsement of various policy initiatives, beliefs about the purpose of prison, and intention to take action in support of prison reform. Findings suggest that reading a passage of an incarcerated individuals’ experience increased self-reported support for prison reform and that an additional nudge that asked participants to put themselves in the shoes of the incarcerated individual to try and understand how they would feel may increase intention to take action in support of prison reform, compared with reading the passage without this additional prompt.

Perspective-Getting (PG) and Perspective-Taking (PT)

The literature on perspective-focused interventions is broad in both its fields of application and manners in which it is implemented. One way to categorize the literature on perspectives is by focusing on the distinction between PT and PG. PT is when an individual imagines what another would feel like by attempting to understand their perspective or point of view. There is a wealth of empirical research studying the impact of PT on empathy for others (Herrera et al., 2018; Skorinko et al., 2014; Swim & Bloodhart, 2015; Todd, Hanko, et al., 2011). PT has been shown to encourage empathy for and reduce prejudice or bias against members of an out-group—a social group that the perspective-taker does not belong to—such as individuals of different racial and/or ethnic identities or nationalities (Bilali & Vollhardt, 2013; Bilewicz, 2009; Müller et al., 2011; Shih et al., 2009; Simonovits et al., 2018; Todd, Bodenhausen, et al., 2011; Todd & Galinsky, 2014) and individuals who have immigrated (Adida et al., 2018; Kalla & Broockman, 2020). For example, Drwecki and colleagues tasked participants with viewing both Black and White individuals’ facial expressions which indicated they were experiencing pain. Participants were told to either make the “best, most accurate” treatment decision or perspective-take (i.e., consider how pain may affect this individual’s life) before making a treatment decision. PT groups demonstrated 55% less anti-Black treatment bias compared with the control group, with no significant evidence of racial bias in pain treatment observed for PT groups (Drwecki et al., 2011). Other applications have demonstrated that PT can increase awareness of intergroup disparities (Todd et al., 2012) and increase donations to causes supporting the well-being of the out-group, regardless of whether the target recipient is in the participant’s in-group or not (Hung & Wyer, 2014).

PG, by contrast, involves actively providing the perspective of another. Like PT, PG interventions have demonstrated success at improving empathy for or understanding of how others feel (Bruneau & Saxe, 2012; Eyal et al., 2018; Zhou et al., 2017), including members of an out-group (Kalla & Broockman, 2021; Okonofua et al., 2020). Audette and colleagues aimed to reduce negative attitudes toward immigrants—specifically Somalis in Kenya. In collaboration with a local advocacy organization, they carried out a public opinion survey and participants were randomly assigned to listen to either one of the recorded perspective-focused narratives or nothing. Participants assigned to listen to a recorded narrative had more empathy for Somalis on average, with the group listening to the narrative describing difficult circumstances faced by refugees reporting greater closeness with Somalis and were more likely to indicate support for keeping the refugee camp open (Audette et al., 2020).

Some studies have found negative effects in certain applications of PT (Epley et al., 2006; Eyal et al., 2018; Hung & Wyer, 2014; Tarrant et al., 2012). One meta-analysis across a variety of experimental contexts, including interpreting facial cues and body language, found that PT decreased an individual’s ability to interpret how another was feeling (Eyal et al., 2018). Epley and colleagues conducted a study in which they recruited undergraduates to bake cookies in groups who were told that the group that made the “best cookies” would win US$100. Baking supplies available to groups were limited, varied in their quality, and needed to be shared among other groups or individuals (i.e., individuals were randomly assigned to either be in competition with those in the group or collaborate with those in their group). They found that among participants who were assigned to compete against others in their group, those assigned to the PT condition, on average, took more resources than the control and took more supplies than what they reported it was fair to take. In fact, within this context, PT led participants to believe that others would be more selfish, leading them to take a higher percentage of the communal resources (Epley et al., 2006). Other studies have found evidence of substantial heterogeneity in the effect of PT on empathy by key factors, such as strong in-group identification; PT may in fact decrease empathy for others within this sub-group (Tarrant et al., 2012).

Many factors likely contribute to the variation in treatment effect estimates of PT interventions. First, previous studies have used a variety of designs and modalities for implementing PT interventions. PT methods range from reading a paragraph or receiving a PT-focused version of instructions for their task (Drwecki et al., 2011; Epley et al., 2006) and engaging in face-to-face dialogue between parties (Bruneau & Saxe, 2012), to answering questions designed to put participants in another’s shoes (Adida et al., 2018). Bilali and Vollhardt (2013) even used priming to induce PT when they conducted a study in Rwanda, where a radio drama had been airing that was designed to promote PT in the context of intergroup conflict. Their PT intervention involved an audio-delivered questionnaire, with the treatment group audio being recorded by the same narrator as the radio drama to illicit a priming effect, which did increase PT (Bilali & Vollhardt, 2013).

A second likely contributor to the variation in findings from PT literature is the variation in contexts applied and information supplied about the out-group or targeted perspective. Todd and Galinsky outline four contexts where PT interventions should be approached with caution: (a) when participants have very fervid in-group identification (see Tarrant et al., 2012), (b) when there are enduring conflicts between the participants and the individual or group of interest, (c) when there is competition between participants and the individual or group of interest (see Epley et al., 2006), and (d) when the perspective that participants are asked to take is of an individual or group that is strongly disliked (Todd & Galinsky, 2014). Some studies prior to their recommendations—for example, Epley’s cookie-baking study—found that PT had a negative impact on empathy for others in the context of competition with out-groups and did not provide information on the perspective of the out-groups (i.e., their thoughts on what percentage of resources would be fair to take). Participants with limited information about another’s perspective may end up relying largely on stereotypes (Gehlbach & Brinkworth, 2012) or egoism (Epley et al., 2006). As a result, interventions should provide participants information from the individual or group they are attempting to take the perspective of—or encourage the participant to seek it out—rather than through their own, independent reasoning while they attempt to put themselves in another’s shoes. In contrast, much of the previously summarized PT literature that found positive effects relied upon providing perspectives to the participant, such as narratives, media, or other sources of information, before applying the PT intervention. In other words, a uniting feature of many of these positive applications of PT is the additional inclusion of PG. In summary, perspective-focused interventions may be most successful when combining PG and PT.

Thus, in this study, I employ PG by providing a narrative description of suggestions on how to deal with quarantine from an individual who is incarcerated to test its impact, compared with information only (IO), in addition to a PT prompt presented before the passage that asks participants to consider how the author of this narrative would feel, compared with PG only. Another area of the perspectives literature that has been studied, though has received limited attention, is the concept of perceived PT, or the feeling that someone is taking our perspective. Over the course of six studies conducted by Goldstein and colleagues, the authors found that engaging in perceived PT—or when a participant believes that someone else has tried to understand their perspective—demonstrated some of the same benefits as PT itself, including increased self–other overlap—a composite measure of closeness, ties, similarity, and other measures—between the participant and an out-group (Goldstein et al., 2014). Thus, this study also aims to test the impact of perceived PT, specifically whether an additional open-ended question asking participants to share their thoughts about COVID-19 and incarceration in the United States improves support for prison reform. Given the personal nature of public safety and its complicated intersection with mass incarceration in the United States, having an opportunity to freely share their thoughts on the pandemic and incarceration may be welcome. Of course, there is also the possibility that asking for an individual’s opinion may center the individual more, which could inhibit their ability to take the perspective of another (Hung & Wyer, 2014). Likely, an individual’s propensity for empathy will be key in moderating the effect of attempting to understand another’s perspective on positive outcomes related to supporting those individuals (Singer & Tusche, 2014).

Importantly, perspective-focused interventions in the prison reform context are only useful to the extent that they may aid in facilitating policy change. Thus, to appropriately address the propensity of perspective-focused interventions to achieve their desired outcome, it is essential to address the influence of power. Extant literature has documented a lack of demonstrable benefit—or even a negative effect—of perspective-focused interventions when the participants perceive themselves to hold substantial power (Galinsky et al., 2006; Ku et al., 2015). Therefore, if our ultimate goal is to change policy, using these interventions in a large sample of individuals who do not wield significant decision-making power in the criminal justice system (e.g., voters, individuals on an email listserv), rather than those at the level of leadership within an organization, is likely the most appropriate application. While PG and PT will not solve complex, societal problems such as systemic racism, anti-poverty sentiment, and mass incarceration, given extant evidence suggests that perspective-focused interventions help encourage empathy for others, these interventions may help bolster support for prison reform, and thus the reason for this study.

The Current Study

I have three primary hypotheses that were a priori specified in the pre-registered analysis plan²:

• Hypothesis 1 (H1): Being assigned to a PG narrative task will improve support for prison reform, compared with being assigned to an IO task.

• Hypothesis 2 (H2): An additional PT prompt bolsters the impact of a PG narrative task in improving support for prison reform.

• Hypothesis 3 (H3): An additional prime of having your viewpoint heard bolsters the impact of a PT intervention in improving support for prison reform.

Support for prison reform is measured by one of three outcome measures, including (a) self-reported support for prison reform initiatives, (b) belief in the rehabilitative purpose of prison, and (c) greater indication of taking action in support of prison reform. There were nine prison reform initiatives, and willingness to support each was measured on a 5-point Likert-type scale from not at all willing to very willing. Each of the nine initiatives are as follows: (a) providing hand sanitizer, masks, and other personal protective equipment to incarcerated individuals in prison; (b) providing hand sanitizer, masks, and other personal protective equipment to staff working in prison; (c) providing free phone calls to incarcerated individuals; (d) increasing use of alternatives to incarceration (e.g., community service, rehabilitative treatment); (e) reclassifying some low-level, non-violent felonies as misdemeanors; (f) early release of elderly incarcerated individuals; (g) early release of incarcerated individuals that are at high-risk for contracting COVID-19; (h) early release of incarcerated individuals who committed a non-violent offense; and (i) early release of incarcerated individuals, regardless of offense type.

To construct the support for prison reform initiatives index, items were summed and scaled from zero to one for easier interpretation of coefficients. Values closer to one indicate greater support for reform. The Cronbach’s alpha coefficient for the support for prison reform initiatives index was .84, indicating strong consistency. A factor analysis was also conducted of the nine items that made up the support for prison reform initiatives index (see Supplementary Appendix 2 for a more in-depth discussion). Three factors were retained, although the third was uninformative. Importantly, the first factor had positive and mostly quite large factor loadings for all items. Thus, the primary analysis proceeds with the index including all nine items; however, Supplementary Appendix 2 includes further results for transparency. The belief in the rehabilitative purpose of prison single survey item measured whether participants felt that the purpose of prison should be rehabilitation, punishment, or both. This item was measured on a 5-point Likert-type scale from totally punishment to totally rehabilitation, with larger scores indicating greater belief in a rehabilitative purpose of prison. Finally, participants were asked three action-oriented items: (a) whether they were interested in receiving more information about prison reform initiatives; (b) whether they would consider making a tax-deductible donation to the Last Prisoner Project, which helps pay for legal/medical care and other services for incarcerated individuals during the COVID-19 pandemic; and (c) whether they would like to sign an open letter to the Department of Justice (DOJ) to support early release for individuals incarcerated in a federal prison to help slow the spread of COVID-19. For the action-oriented items outcome, these three items were summed, ranging from zero to three, where zero indicates no intention to take action in support of prison reform, and three indicates intention to take action in support of all three initiatives. The Cronbach’s alpha coefficient for the action-oriented items outcome was .72, slightly higher than the .70 standard for acceptable consistency (Tavakol & Dennick, 2011).

Method

A total of 2,662 participants were initially recruited from Amazon’s MTurk platform.³ Workers were paid US$2.00 for completing the 10-min survey in its entirety.⁴ Workers were required to pass an attention check before completing the survey. Of 2,662 initial entrants, 2,318 (87.1%) workers passed the attention check. The final sample size, after removing a small number of duplicate responses (i.e., multiple instances of the same worker identifier captured through embedded survey data) and those that failed the attention check, was 2,299. All participants were first asked a series of demographic questions, questions about political preference and partisanship, measures of empathic concern and PT ability,⁵ and their experience with COVID-19. The experience of the virus outbreak questions included whether they were employed during the pandemic, or if they lost their job due to being furloughed or laid off, as well as the aspects of their life that have become more stressful, including handling conflicts with people in or outside of the home, dealing with financial stress, finding time/space and privacy, and having enough time to supervise children, if applicable.⁶ Then participants were randomly assigned to one of the four conditions, as detailed below, with the intended percentage of individuals to be allocated to this condition in parentheses:

1. Information only—IO (30%): The information task asked participants to read a 102-word paragraph that detailed the impact of COVID-19 on incarcerated individuals in neutral language (see Supplementary Appendix 1 for exact language). Following the informational paragraph, participants were asked a question that gauged their comprehension of the information presented. Participants assigned to the IO condition received just the information task.

2. Information, followed by the PG narrative with control prompt—PGN-C (30%): Participants in the PGN-C condition received the same information task as the IO condition, followed by a PG narrative. The 139-word narrative provides advice for dealing with quarantine in the time of COVID-19 from the perspective of someone who has been incarcerated for 25 years. This narrative was written by Jerry Metcalf, a writer for the Marshall Project, who was incarcerated at the time the narrative was written (Metcalf, 2020). Following the narrative, participants are asked how easy it was to consider what someone in that situation might be feeling. Individuals in the PGN-C condition received a standard prompt to read the narrative and answer the question that followed (see Supplementary Appendix 1 for exact language).

3. Information, followed by the PG narrative with PT prompt—PGN-PT (20%): Individuals assigned to the PGN-PT condition went through the same tasks as the PGN-C condition; however, they received an additional prompt, designed to encourage PT: “While reading, please mentally walk in the shoes of someone who is incarcerated in these conditions. You should concentrate on the way they would feel about what has happened by imagining you are actually them. As you read, try to put yourselves in the author’s shoes.” The purpose of this prompt was the provide an extra nudge to encourage participants to take the perspective of an incarcerated individual living in confinement.

4. Information, followed by an open-text question for sharing thoughts (perceived PT), and finally the PG narrative with PT prompt—PGN-PPT (20%): Individuals assigned to this condition had the same tasks and text as those in the PGN-PT condition; however, between the information and narrative tasks, participants were given the opportunity to share their thoughts on COVID-19 and incarceration in the United States in an open-text response (see Supplementary Appendix 1 for exact question).

Ordinary least squares (OLS) models with heteroskedasticity-robust standard errors were utilized to test each hypothesis using the Westfall–Young method of estimating adjusted p values with free step-down resampling for multiple comparisons (Westfall & Young, 1993). For H1, testing the impact of PG compared with IO utilized a comparison between the combined PG groups—PGN-C, PGN-PT, and PGN-PPT—and the IO group. Comparisons made in H2 are for the purpose of testing whether the additional PT prompt improves support for prison reform. Therefore, the IO condition was not included as no PG narrative was present for this condition. The treated group for this comparison is the combined PT groups (PGN-PT and PGN-PPT), compared with the control of PG only (PGN-C). Finally, H3 tests the impact of an additional perceived PT task in combination with PT, therefore, the comparison is the PGN-PPT condition versus the PGN-PT condition (with IO and PGN-C excluded as no PT is included in these treatment arms).

After reading the passage(s) corresponding with their condition, participants were presented with each group of outcomes previously described—the support for prison reform initiatives, their belief about the purpose of prison, and the action-oriented outcomes (getting more information about prison reform, donating to a charity supporting legal, medical, and other services for incarcerated individuals, and signing an open letter to the DOJ to release incarcerated individuals from prison). The full survey is available on this study’s OSF webpage.⁷

Results

Table 1 includes descriptive statistics of the 2,229 survey participants.⁸ Of the 2,299 participants, most identified as female (54.2%) and less than 2% identified as non-binary, transgendered, another gender identity, or preferred not to state their gender. The racial and ethnic identities of participants was largely White (77.2%), with 13.7% identifying as Black, 7.8% identifying as Asian, and less than 2% each identifying as Native Hawaiian/Pacific Islander or American Indian/Alaskan Native. Slightly less than 10% of participants identified as Hispanic or Latinx. Compared with the U.S. population,⁹ the survey sample was more left-leaning, with almost half (45.8%) of participants identifying as Democrat and more than 50% identifying as liberal. The average age was about 37 years old and more than two thirds (66.9%) were employed at the time of the survey.

Table 1.

Descriptive Statistics of the Survey Participants.

Variable	Level	Frequency	Percentage
Treatment arm	IO	692	30.10
	PGN-C	690	30.01
	PGN-PT	460	20.01
	PGN-PPT	457	19.88
Gendera	Female	1,245	54.15
	Male	1,018	44.28
	Non-binary	20	0.87
	Other or prefer not to say	16	0.70
Raceb	American Indian or Alaskan Native	44	1.91
	Asian	180	7.83
	Black	315	13.70
	Native Hawaiian or Pacific Islander	10	0.43
	Other	51	2.22
	White	1,775	77.21
	Prefer not to say	27	1.17
Ethnicity	Hispanic or Latinx/Latino/Latina	225	9.79
Employed?	No, no job prior to COVID	459	19.97
	No, recently laid off/furloughed	301	13.09
	Yes	1,539	66.94
Partisanship	Democrat	1,053	45.80
	Independent	610	26.53
	Other or Third Party	105	4.57
	Republican	531	23.10
Political preferencec	Very conservative	173	7.53
	Conservative	274	11.92
	Slightly conservative	226	9.83
	Moderate	460	20.01
	Slightly liberal	309	13.44
	Liberal	525	22.84
	Very liberal	270	11.74
	Don’t know or no preference	62	2.70
Variable	Mean (SD)	Minimum	Maximum
Age	36.62 (11.50)	18	77
Perspective-taking	19.17 (4.72)	0	28
Empathic concern	20.12 (5.22)	0	28
COVID stress index	4.42 (2.27)	1	10

Note. IO = information only; PGN-C = perspective-getting narrative with control prompt; PGN-PT = perspective-getting narrative with perspective-taking prompt; PGN-PPT = perceived perspective-taking open-ended task + perspective-getting narrative with perspective-taking prompt.

^aTransgender female and Transgender male were also included in other, but given very small cell frequencies, were not reported for purpose of protecting confidentiality. For estimation purposes (i.e., to have large enough cell sizes), non-binary and other or prefer not to say were included in the same level.

^bFor estimation purposes (i.e., to have large enough cell sizes), American Indian or Alaskan Native and Native Hawaiian or Pacific Islander were included in the same level as individuals of a race other than those listed and those that preferred not to state their race.

^cFor estimation purposes, moderate and don’t know or no preference were combined.

In terms of the baseline covariates, the average perspective-Taking and Empathic Concern subscale scores from the Interpersonal Reactivity Index (IRI) were in line with similar measures reported in previous literature (e.g., Yarnold et al., 1996). Interestingly, participants reported slightly less stress related to or because of COVID. The average COVID Stress Index score was 4.42, with the index representing the average of six items on a 10-point Likert-type scale.¹⁰ Questions asked about the extent to which participants felt more stress as a result of COVID across a variety of factors, such as being able to afford essentials, experiencing personal conflict with others in or outside of the home, having enough time to supervise their children, and so on. Understanding the level of stress that individuals are experiencing related to the COVID-19 pandemic is important AS those that have been especially affected by the pandemic may respond differently to perspective-focused interventions.

As shown in Table 2, covariates are mostly balanced across treatment condition. Each column represents test statistics for the comparisons made within hypothesis: H1, H2, and H3. For the comparisons made in H1 (all three PGN conditions compared with IO) and H2 (both PGN-PT/PPT conditions compared with PGN-C), there was one significant difference between the treated conditions and the control condition. Specifically, for the comparisons made in H1, the treated group (PGN conditions) had a significantly higher proportion of individuals identifying as Native Hawaiian and/or Pacific Islander and the treated group for comparisons made in H2 (PGN-PPT and PGN-PT) had a significantly higher proportion of individuals identifying as a race other than those listed. These imbalances were small and the frequencies in these two racial identities were low, especially for comparisons made in H1. Notably, the comparisons made in H3 (PGN-PPT vs. PGN-PT) had a few factors that were significantly different, and one considerably so: A substantially higher percentage of the PGN-PPT group identified as Hispanic or Latinx (12.91%) compared with the PGN-PT group (7.61%).

Table 2.

Balance Table.

Variable	${\chi }_{\text{H}1}^2$	${\chi }_{\text{H}2}^2$	${\chi }_{\text{H}3}^2$
Gender	6.43†	4.23	1.13
American Indian or Alaska Native	0.17	0.39	3.33†
Asian	0.67	1.81	0.07
Black	<0.01	3.34	1.32
Native Hawaiian or Pacific Islander	4.33*	0.20	0.21
Other race	0.04	4.25*	1.19
White	1.77	<0.01	1.02
Hispanic/Latinx	2.21	0.05	7.00**
Partisanship	2.28	3.33	0.93
Political preference	1.72	5.73	10.18
Variable	t H1	t H2	t H3
Age	−0.38	1.56	2.25*
Perspective-taking	−0.63	1.70	1.13
Empathic concern	−0.50	1.81	0.34
COVID stress index	0.05	0.09	−0.46

Note. Subscripts H1 to H3 indicate the hypothesis being tested. Due to expected cell frequencies less than five, the Fisher’s exact test was used for the binary indicator for identification as a Native Hawaiian or Pacific Islander.

^†p < .1.

^*p < .05.

^**p < .01.

Much smaller was the age difference, with the PGN-PT group a little less than 2 years older, on average, compared with the PGN-PPT group. Given that COVID-19 and mass incarceration has had disparate impacts on people of color, especially those identifying as Black, Latinx, and/or Native American, largely as a result of discrimination and systemic racism (Gray et al., 2020), the amount of stress they may have been subjected to during the pandemic could be in part due to these inequities. In terms of stress level, findings show higher stress for those identifying as Latinx (5.42), Black (5.08), and/or American Indian (4.56), compared with the overall mean (4.42) and White participants (4.33). As the level of stress individuals are experiencing because of COVID-19 may affect their propensity for engaging with tasks, this is important to note.

Turning to the primary outcomes, there was variation in unadjusted outcome means across the four treatment arms (see Figure 1). The three PG arms (PGN-C, PGN-PT, and PGN-PPT) had higher average policy support, compared with the IO group. The average of purpose of prison outcome was fairly consistent across treatment arms. The average number of action-oriented items endorsed varied across treatment arms with the IO group having the lowest average and the PGN-PT group having the highest.

Figure 1.

Mean outcomes by treatment group.

Note. IO = information only; PGN-C = perspective-getting narrative with control prompt; PGN-PT = perspective-getting narrative with perspective-taking prompt; PGN-PPT = perceived perspective-taking open-ended task + perspective-getting narrative with perspective-taking prompt.

Main Hypothesis Testing

Each hypothesis was tested via an OLS model with heteroskedasticity-robust standard errors. To correct for multiple comparisons, the Westfall and Young method of estimating adjusted p values with free step-down resampling was applied (Westfall & Young, 1993). Table 3 reports the average treatment effect (ATE) estimates for each hypothesis (H1, H2, and H3) on each outcome (sum of self-reported support for prison reform initiatives scaled from zero to one, purpose of prison, and sum of action-oriented outcomes), including both unadjusted and Westfall–Young p values.

Table 3.

Outcome Model Comparisons by Hypothesis.

Hypothesis	Outcome	Model	Coefficient	Robust SE	p value (unadjusted)	p value (Westfall–Young)
Hypothesis 1 (PGN vs. info alone)	Policy support sum (scaled)	T	0.030	0.009	<.001***	<.001***
		T + all controls	0.024	0.007	<.001**	.007**
	Purpose of Prison	T	0.031	0.046	.499	.491
		T + all controls	0.016	0.041	.694	.705
	Action-oriented sum	T	0.098	0.052	.059	.110
		T + all controls	0.060	0.047	.197	.354
Hypothesis 2 (PT vs. PG)	Policy support sum (scaled)	T	0.007	0.009	.420	.660
		T + all controls	0.008	0.008	.315	.522
	Purpose of prison	T	0.031	0.050	.540	.660
		T + all controls	0.019	0.045	.673	.670
	Action-oriented sum	T	0.083	0.058	.155	.382
		T + all controls	0.101	0.052	.051	.143
Hypothesis 3 (PPT vs. PT)	Policy support sum (scaled)	T	−0.002	0.012	.856	.870
		T + all controls	−0.009	0.010	.389	.615
	Purpose of prison	T	0.054	0.066	.413	.612
		T + all controls	0.026	0.059	.657	.690
	Action-oriented sum	T	−0.093	0.078	.232	.493
		T + all controls	−0.142	0.068	.036*	.110

Note. All controls include empathy subscales, COVID questions, demographics, and politics. SE = standard error; PGN = perspective-getting narrative; PT = perspective-taking prompt; PPT = perceived perspective-taking open-ended task + perspective-getting narrative with perspective-taking prompt.

^*p < .05.

^**p < .01.

^***p < .001.

For H1, I find a significant, positive impact of PG on support for policy reform initiatives. PG increased self-reported support for prison reform policies by 0.03 points on a zero-to-one index. In other words, receiving the PG narrative resulted in a 3% increase in support for prison reform initiatives, compared with receiving information alone. Including covariates reduces this estimate slightly, although 95% confidence intervals suggest possible effects between 1.2% and 4.8%. Supplementary Appendix 2 provides a comparison of the different specifications for this index based on results of the factor analysis (i.e., focusing on COVID-19 policies specifically as well as non-COVID-19 policies, with removal of the early release of all individuals item). Results were robust, although there was no significant difference between the PGN and IO groups for the COVID-19-related policies, likely due to their already overwhelming support in the sample (more than 69% were very willing to support providing PPE (personal protective equipment) for both incarcerated individuals and corrections staff). While the belief that purpose of prison was rehabilitation over punishment and sum of action-oriented outcomes were greater for the PGN groups compared with the IO group, these differences were not statistically significant.

For H2, I do not find substantial evidence of an additional benefit of a PT prompt in combination with PG, comparing the PGN-PT conditions (PGN-PT and PGN-PPT) with the PGN-C condition. Estimates of the ATE for the policy support outcome were quite close to zero. However, for the purpose of prison and sum of action-oriented outcome models, the estimates of the ATE were positive, but none were statistically significant.

Contrary to my third hypothesis, perceived PT—providing one’s thoughts and opinions about COVID-19 and incarceration in the United States—reduced the number of intended actions in support of prison reform. The only significant models for this outcome, however, are based on the unadjusted p values and not the multiple comparisons corrected Westfall–Young p values. Still, these results suggest that asking participants to provide their perspective in addition to a PT intervention may have a negative impact on the intention to act in support of prison reform, compared with PT alone. Perceived PT had no impact on support for prison reform relative to PT alone. I find a small effect of perceived PT, compared with PT alone, on beliefs about the purpose of prison, although this difference is not statistically significant.

In terms of the magnitude of these effects, Table 4 includes the Cohen’s d effect sizes for each outcome and the comparison groups. The largest effect size was for the combined three PG groups (PGN-C, PGN-PT, and PGN-PPT) versus IO at 0.162 for self-reported support for prison reform. Although small, these effect sizes are consistent with results reported in extant literature on perspective-focused interventions (Kalla & Broockman, 2021).

Table 4.

Cohen’s d Effect Sizes.

Outcome	Cohen’s d	Hypothesis	Comparison
Policy support sum (scaled)	0.162	1	PG vs. IO
Purpose of prison	0.031
Action-oriented sum	0.085
Policy support sum (scaled)	0.039	2	PT vs. PG
Purpose of prison	0.031
Action-oriented sum	0.071
Policy support sum (scaled)	−0.011	3	PPT vs. PT
Purpose of prison	0.054
Action-oriented sum	−0.079

Note. IO = information only; PGN = perspective-getting narrative; PT = perspective-taking prompt; PPT = perceived perspective-taking open-ended task + perspective-getting narrative with perspective-taking prompt.

In an exploratory analysis, I also compared the PGN-PT condition with the PGN-C group on the sum of action-oriented items outcomes, given the above results suggesting a negative effect of the perceived PT task (see Table 5). Comparing PGN-PT with PGN-C directly, I find a significant difference for the sum of action-oriented items outcome, with an estimated ATE of 0.17 and 95% confidence interval of [0.04, 0.29], indicating that on average, individuals who received the PG narrative with the additional PT prompt (but no perceived PT) intended to take more action in support of prison reform on average, compared with those that just received the PGN-C. However, in models without COVID-related covariates, this difference was smaller in magnitude and not statistically significant. This suggests an avenue for future research and exploration.

Table 5.

Outcome Models—Difference Between PT and PG (No PPT Arm).

Hypothesis	Outcome	Model	Coefficient	Robust SE	p value (unadjusted)	p value (Westfall–Young)
PT vs. PG (not including PPT, IO)	Policy support sum (scaled)	T	0.008	0.011	.440	.671
		T + all controls	0.013	0.009	.167	.304
	Purpose of prison	T	0.004	0.061	.953	.964
		T + all controls	0.013	0.055	.817	.821
	Action-oriented sum	T	0.129	0.070	.067	.177
		T + all controls	0.168	0.062	.007**	.022*

Note. All controls include empathy subscales, COVID questions, demographics, and politics. SE = standard error; PT = perspective-taking prompt; PG = perspective-getting; PPT = perceived perspective-taking open-ended task + perspective-getting narrative with perspective-taking prompt; IO = information only.

^*p < .05.

^**p < .01.

^***p < .001.

Supplementary Analyses

As specified in the pre-registered analysis plan, I conducted supplementary analyses to examine whether the effect of perspective-focused interventions on policy support differs by type of reform. My primary analysis found varying magnitudes of the effect of PG, compared with information alone. In terms of statistically significant findings, three of the nine initiatives—providing free phone calls to incarcerated individuals, increasing the use of alternatives to incarceration, and early release of incarcerated individuals who committed a non-violent offense—had significantly greater support, on average, for PGN groups compared with the IO group. Support for the other proposed reforms, other than providing PPE for staff, was somewhat greater among participants in the PGN conditions (PGN-C, PGN-PT, and PGN-PPT), relative to those assigned to IO, although observed differences were often small.

Discussion

In this study, I found that PG increased empathy for incarcerated individuals, specifically bolstering participants’ self-reported support for prison reform initiatives, compared with information alone. Especially considering its ease of implementation and low cost, even finding small effect sizes in this context are promising and worthy of future research that can address the extent to which perspective-focused messaging can be scaled up and applied to contexts where we can measure the impact of the intervention on action rather than intention to take action. The population of interest in this study was the general public, and therefore, seeking opportunities to validate these findings in larger samples of Americans¹¹ and samples outside of the context of a survey experiment is strongly advised. For example, campaigns for prison reform–oriented propositions within local and state governments could utilize a perspective-focused intervention through mail-based advertisement to registered voters shortly before elections to measure its impact on support for the proposition.

Outside of the context of public support, it is also crucial to consider the support for these reforms among system stakeholders, in particular, those on the frontlines. Given that these stakeholders are typically those in charge of implementing reforms, bolstering their support of these initiatives is vital to their success. Therefore, it may be worth encouraging PT and PG among stakeholders to better understand the perspectives of those who corrections policy most profoundly affects—those who are incarcerated within correctional institutions. PT ability or interventions have not been widely studied within the field of corrections; however, there has been some research on PT in the context of the criminal justice system more broadly. Specifically, one study of Swedish National Police¹² found that longer tenured officers tended to measure lower on PT scales, relative to officers who were earlier on in the career (Inzunza et al., 2019). While there are reasons to believe that what attracts individuals to a career in policing may change over time, this study also highlights the role of institutionalization and an opportunity to incorporate regular training or opportunities to exercise PT and PG within the criminal justice system to help frontline officers better connect with those they serve. Within corrections specifically, using perspective-focused interventions to improve support of reform among staff may be impactful, although employees’ perceptions of their level of power (Galinsky et al., 2006; Ku et al., 2015), among other contextual factors, should be considered quite carefully before deciding to implement these interventions. Extant literature has identified factors that are associated with officer support of rehabilitation and that the experiences of frontline staff should be taken into consideration when implementing corrections policy reform. Specifically, the higher the perceived support from and competency of supervisors, percentage of incarcerated individuals involved in high-quality programming, and institutional response to violence, the greater the support for a rehabilitative purpose of prison (Lerman & Harney, 2019). In conjunction with Todd and Galinsky’s (2014) findings suggesting the potential for PT to backfire in contexts where there is persistent conflict between the target group and participants, immediate application of perspective-focused interventions among correctional officers may not be advised. Addressing pressing conflicts, such as institutional violence exposure between incarcerated individuals and staff (Lerman et al., 2021), or sources of conflict, such as insufficient access to mental health services or other resources (Lerman et al., 2021), should likely precede any applications of perspective-focused interventions. In other words, acknowledging institutional constraints and correctional employee needs is likely an important component to pair with or precede a perspective-focused intervention aimed at improving support for prison reform, and could be crucial to improving well-being and safety outcomes for the individuals under their care.

In addition, while the evidence was discovered in an exploratory analysis, it is promising that an additional PT prompt, in combination with PG, may increase the level of intended action in support of prison reform. This finding somewhat conflicts with recent research examining the additional impact of PT. Kalla and Broockman studied the impact of different narratives, including a PG narrative and an analogic PT narrative, or PT that asks participants to reflect on a similar experience from their own life. They found that the addition of an analogic PT narrative did not do significantly better than simply using PG alone (Kalla & Broockman, 2021). An important difference between their PT narrative and the PT prompt utilized in the current study is that the prompt was not analogic given the nature of the perspective provided in this study. Specifically, for participants who have never been incarcerated, they will not be able to understand what incarceration feels like, and although their COVID-19 experience may help them become more cognizant of the difficult circumstances of incarceration, this will not equate to a personal understanding of what it means to be incarcerated. Therefore, the PT prompt in this study asked participants to focus only on how another may feel rather than consider an event from their own experience. Future studies should pursue testing the additional benefit of PT in combination with PG more broadly—across different contexts, with a larger sample, and designed so that direct action and endorsement of reform initiatives can be accurately and appropriately measured.

Finally, I found evidence that perceived PT did not bolster the impact of PT; in fact, there was evidence to suggest that it lowered participants’ intention to take action, on average, relative to those not receiving the additional perceived PT prompt. There are several possible reasons for which this may be the case. First, the task itself could be insufficient at encouraging participants to share their thoughts or feel that their thoughts have been heard. In terms of engagement with the task, on average, participants in the PGN-PPT group wrote a response that was about 229 characters long, with the median response being 172 characters long, or approximately two sentences. The 25th and 75th percentiles for number of characters were 88 and 314, respectively, suggesting that a sentence or three was a typical response length for participants in this group. While this level of engagement is in line with expectations, it is certainly possible that this task needed to be longer. In addition, the venue of an online survey may not properly provide individuals with the feeling that their opinion matters, or their perspective is being considered, given previous literature demonstrating a positive effect was through a simulated conversation, relying upon chat bots (Goldstein et al., 2014). It could also have been the case that these individuals did feel that their perspectives were being taken and that in the context of a global pandemic where individuals are overwhelmed, anxious, and exhausted, that providing the opportunity to focus on their own experience and opinions detracted from the potential benefit of taking the perspective of another. While offering opportunities for individuals to feel heard and understood is valuable and important, it may be best to consider context, medium, and time when deciding how—or if—one implements perceived PT. Although future studies are warranted here, these findings demonstrate a need to proceed with caution in the application of perceived PT interventions with the goal of increasing empathy for others.

The findings from this experiment contribute meaningfully to multiple literatures, including the literature on perspective-focused interventions. To the author’s knowledge, perspective-focused interventions have not been applied to the context of attitudes and beliefs about the criminal justice system. Prior to this study, there was also limited to no evidence of the impact of non-analogic PT in combination with PG, compared with relying upon PG alone. More broadly outside of this study’s contribution to the perspective-focused literature, this inquiry contributes to evidence on low-cost ways of encouraging support reforms and policy initiatives among the general public.

As with all empirical research, this study has a few limitations. First, while the PG narrative significantly increased self-reported support for prison reform policies, it is unclear how this might translate into real behavior, such as vote choice on proposed prison reform legislation. Similarly, intention to take action is different than actually taking action, and the extent to which a PT prompt into addition to a PG narrative can increase tangible action is not certain. However, there does seem to be promise: A recent study found that in the context of organ donation, PT through reciprocal altruism (“If you needed a transplant, would you have one?”) did increase organ donor registration (Robitaille et al., 2021). Future studies should test perspective-focused strategies in the field to determine if these low-cost interventions can yield benefits in actuality for prison reform (and on a larger scale). The strength of the PT prompt is also key to interpreting these findings appropriately. From previous literature, we know that the frequency and medium matter (Herrera et al., 2018; Kalla & Broockman, 2020). The text-only treatment and one-time intervention make it likely that the effect is not durable. Future studies of perspective-focused interventions’ impact on support for criminal justice policy reform can build upon this work by testing repeated messaging, for example, randomizing not only the intervention but also its duration by varying how long the participants are sent messages, and similar to work from Kalla and Broockman (2021), testing conversation-based versions of perspective-focused interventions.

Ultimately, our ability as a society to bridge gaps between one another is vital to building a more equitable society. As Rios and colleagues (2003) describe, their social PT process involves “… incorporating awareness, analysis, and advocacy” (p. 11). Feedback from the survey participants, delivered as comments after completing the MTurk task, certainly seem to suggest a similar process for some participants. The PG narrative, as well as information, can increase awareness of the horrendous conditions of incarceration—especially during a global pandemic. Some participants noted that they “[were] not aware conditions were so bad in jails,” that “it was … a part of the pandemic that [they] hadn’t really thought about … something the media doesn’t really cover or talk about,” or that their experience of the survey was “Very eye opening to a topic [they] haven’t thought about before!” Perspective-focused interventions may help encourage thoughtful scrutiny of how the United States utilizes incarceration, especially when considering the humanity of those who are subjected to these conditions. One participant commented that the survey “Really challenged [them] with some interesting perspective.” This study demonstrates the promise of perspective-focused interventions and hope that they can build further momentum for reforming the justice system through sparking interest and passion to learn or do more, as one participant noted about taking action to support prison reform noted: “[I] am glad I can do something to help in some small way.”

Biography

Jessie Harney is a graduate student researcher at The People Lab and PhD candidate at the Goldman School of Public Policy at the University of California, Berkeley. Her research interests are in criminal justice system reform with a specific focus on mental health.