Personal Narratives Reduce Negative Attitudes Toward Immigrant Outgroups: Evidence From Kenya

Abstract

Public opposition constitutes an important barrier to refugee and migrant integration in many parts of the world. As a result, scholars have tested a variety of interventions to counteract negative views. We add to the literature on narrative strategies by demonstrating that a light-touch approach—listening to first-person personal narratives from members of an “immigrant outgroup”—can be effective even among those who hold more negative dispositions. We record two personal narratives developed by members of the Somali community in Kenya that highlight: (1) refugee hardships and (2) shared opposition to terrorism. Experimental data from a representative survey in Nairobi show that both treatments have positive effects on intergroup and policy attitudes. The effects are generally as large among those who have more negative predispositions toward Somalis. This approach holds promise as a low-cost, scalable strategy for advocacy organizations seeking to shift attitudes toward immigrant outgroups.

Against the backdrop of rising global migration (UNHCR, 2024), a vast scholarship examines the sources of anti-immigrant attitudes (Hainmueller & Hopkins, 2014). Recently, scholars have started investigating interventions aimed at ameliorating such attitudes through door-to-door canvassing (Kalla & Broockman, 2020, 2023), vicarious intergroup contact (Alrababa'h et al., 2021; Weiss et al., 2023), perspective-taking (Adida et al., 2018; Alan et al., 2021; Larson & Lewis, 2025), priming superordinate identities (Rosenzweig & Zhou, 2021), and correcting misperceptions with statistics or facts (Hopkins et al., 2019). One widely-cited challenge for such interventions is motivated reasoning—the tendency for individuals to reject or dismiss information or other attempts at persuasion that conflict with prior views, particularly among the most ardent opponents (Druckman, 2022; Jost et al., 2022; Taber & Lodge, 2006).

In this study, we investigate whether listening to a first-person personal narrative from members of an immigrant outgroup is effective in reducing negative intergroup and policy attitudes. By immigrant outgroups, we mean marginalized communities that include refugees and immigrants, alongside longtime citizens from the same group, who are often perceived as immigrants or outsiders by other citizens. By first-person personal narrative, we mean a personal account that conveys experiences and/or perspectives from a member of the target group in their own voice. To develop this intervention, we partnered with a Somali advocacy organization in Kenya to generate and record two audio treatments that convey the perspectives of distinct sub-groups. The first treatment focuses on refugees from Somalia who have fled violence and instability in that country. The second relates to the experiences of long-term Kenyan citizens of Somali descent, who are often viewed as “foreign” despite having part of their traditional homeland within Kenya’s borders. We then examine whether listening to these narrative treatments over headphones has the power to reduce negative intergroup and policy attitudes towards Somalis among other Kenyans.

While one might question whether narrative interventions can nudge attitudes regarding a controversial topic, there are compelling theoretical reasons why this approach may be effective. According to transport theory from the communications literature, personal narratives may have the power to circumvent motivated reasoning, reducing counter-arguing and increasing message receptivity and attitudinal change (Green & Brock, 2000). Additionally, models of exemplar-based attitude formation from psychology suggest that exposure to positive information about individual outgroup members can lead respondents to update views on the overall outgroup (Gawronski & Bodenhausen, 2006; Paolini & McIntyre, 2019; Zillmann & Brosius, 2012).

Kenya offers an important context in which to test this intervention, given its status as one of the largest refugee-hosting countries in the world. Somalis, who constitute the largest single refugee group in Kenya, are a marginalized community comprised of both longtime citizens and more recent refugees. Public opinion towards the Somali community, and policies aimed to benefit it, is overwhelmingly negative due to the rise in terrorism perpetrated in Kenya by Al Shabaab, a Somalia-based insurgent group (IRC, 2018). Negative media portrayals contribute to the public association of Somalis with terrorism. Our own analysis of media coverage in the year prior to this study finds that over 75% of newspaper articles regarding Somalis mention violence and security threats. The negative association of Somalis with insecurity has direct implications for policy debates, serving as the pretext for the Government’s repeated threats to close the Daadab refugee camp in northern Kenya, which mainly houses Somali refugees.

In contrast to studies in which researchers craft treatment scripts (Druckman, 2022), we asked members of a local Somali advocacy group in Kenya to create the narratives. Specifically, we asked the group to design content that they believed was missing from the media that could potentially counteract negative attitudes toward their community. The first treatment, recorded by a Somali refugee, emphasizes the hardships refugees encounter in leaving their home country, along with expressions of gratitude to the Kenyan public. The second, recorded by a Somali Kenyan (i.e., an ethnic-Somali Kenyan citizen), conveys shared opposition to Islamic terrorism in Kenya.

We test the narratives (relative to a control group) using a survey experiment embedded in a large household survey (n = 1,112) ahead of the 2017 Kenyan elections in Nairobi and find that both treatments affect a range of intergroup and policy attitudes. We observe 6 and 9 percentage point increases on composite indices of outcome measures for the refugee hardship and anti-terror narratives, respectively, and large and substantively meaningful changes on several component items. For example, the refugee hardship narrative increased the mean number of refugees deemed eligible for citizenship by 7 percentage points and increased support for keeping Dadaab open by 17 points. The anti-terror narrative increased perceptions of Somalis as peaceful by 10 points and led to a 14-point increase in the view that Somali Kenyans belong in Kenya. We also find that treatment effects are in most instances as large among respondents who are more negatively predisposed to Somalis, defined as those who (1) implicitly associate Somalis with violence, (2) view Somalis as a labor market threat, (3) perceive Somalis as culturally dissimilar, and (4) have less contact with Somalis.

These results relate to the existing scholarship on narrative interventions aimed at reducing prejudice towards immigrants in three ways. First, evidence in favor of narrative approaches mainly comes from studies in which narratives have been used in conjunction with other complementary strategies and/or delivered through interpersonal exchanges (Alan et al., 2021; Harney, 2023; Kalla & Broockman, 2020, 2023; Larson & Lewis, 2025; Wojcieszak & Kim, 2016). The findings in this study suggest that “light-touch” narrative interventions delivered without interpersonal exchange or other persuasive components can be effective, bolstering the rationale for an approach that has the potential to be scaled by advocacy groups at low cost through social or traditional media (also see Adida et al. (2025)). Moreover, our findings contrast with other recent studies that aim to isolate the effects of narrative treatments using survey-experimental methods and find more mixed results (e.g., Bandiera et al., 2026; Gubler et al., 2022; Kalla & Broockman, 2023). The discussion section addresses these differences and their implications for future research.

Second, this study investigates heterogeneous effects based on prior dispositions toward the outgroup, which has generally not been the focus of prior work. Our findings are noteworthy in that the narratives in this study are in most instances as effective among those who hold more negative views of the outgroup. This finding lends support to the theory that narrative approaches can increase message receptivity among individuals who may be inclined to discount information that conflicts with prior views. Studying heterogeneity is particularly important when examining interventions designed to shift attitudes regarding immigrants and refugees, since advocacy groups typically target their efforts toward those who oppose immigration. Given that treatments designed to increase empathy for immigrants run the risk of “preaching to the choir” (i.e., increasing positive views mainly among those who are already supportive (Gubler et al., 2022)), we suggest that future studies prioritize exploring heterogeneity based on prior dispositions and specifically whether interventions produce shifts among those who oppose immigration.

Third, we demonstrate the value of collaborating with local communities to develop treatments featuring content generated by members of the group. It is important to involve local groups in the design and production of research initiatives—doing so can improve construct and ecological validity, maximizing the likelihood that interventions will work (Michelitch, 2018; Thachil, 2018). We attribute the efficacy of the treatments in this study in part to their presentation as first-person accounts based on deep understanding of the local context by members of the Somali community.

Attitudes towards Immigrant Outgroups and Narrative Interventions

An extensive literature examines opposition to immigrant outgroups in the Global North, and more recently, the Global South. ¹ Negative views are often rooted in misperceptions about the size of immigrant communities, their characteristics, or the costs of providing government services—beliefs that underlie perceived cultural, economic, or security threats (Hainmueller & Hopkins, 2014). Given that host-country citizens typically have little direct contact with members of immigrant outgroups, misperceptions may be fueled by the media, where negative portrayals often predominate (Brader et al., 2008), and may be compounded by scapegoating from political leaders (Whitaker & Giersch, 2015). Particularly relevant to this study, prior research shows that opposition among host-country nationals is especially acute for Muslims in Christian-majority countries due to a perceived association with Islamic terrorism globally (Creighton & Jamal, 2015; Merolla & Zechmeister, 2009).

In this study, we test whether first-person personal narratives in the form of a light-touch informational intervention can ameliorate negative intergroup and policy attitudes. This approach aims to shift attitudes and behaviors by presenting information to an audience without the addition of other intervention elements (e.g., interpersonal dialogue or perspective-taking exercises) commonly included in canvassing and other related interventions (Kalla & Broockman, 2020, 2023; Larson & Lewis, 2025). Light-touch interventions are of interest because they have the potential to be scaled over mass media (e.g., through social media, podcasts, radio, or TV), making them potentially useful for advocacy groups seeking to sway mass attitudes.

The first generation of light-touch informational interventions focused on providing data and statistics designed to correct misperceptions about immigrant outgroups, based on the observation that negative views often stem from inaccurate beliefs about the size, cost, or other characteristics of refugee and migrant groups. This approach produced limited effects, likely because the stability of immigration views makes them resistant to change and because motivated reasoning leads individuals to discount or reject information that conflicts with priors (Druckman, 2022; Hopkins et al., 2019; Taber & Lodge, 2006). ²

This study joins a second approach, borrowed from the communications literature, that focuses on personal narratives as a strategy for shifting attitudes and behaviors. Personal narratives are communicative devices that relate personal experiences, testimonials, and perspectives (Green & Brock, 2000). Communications scholars have long regarded narratives as an effective strategy for persuasion (Braverman, 2008; Green & Brock, 2000, 2002; Krause & Rucker, 2020). A wide range of “edutainment” initiatives—in which persuasive messages in narrative form are embedded in movies, television shows, radio dramas, or other mass media channels—have been shown to be effective in international public health and development programming (Grady et al., 2021; Rahmani et al., 2025).

Personal narratives are thought to enhance persuasion through several related channels (Moyer-Gusé & Nabi, 2010). First, building on transport theory, the literature proposes that when individuals are immersed or “transported” into others’ experiences and perspectives, they are less likely to counter-argue and will therefore be more receptive to persuasion (Green & Brock, 2000; Nabi & Green, 2015; Van Laer et al., 2013). Narrative immersion is thought to shift cognitive effort away from message-countering thoughts (De Wit et al., 2008; Krakow et al., 2018; Wojcieszak & Kim, 2016). Narratives may also enhance believability because it may be difficult to discount or reject an individual’s personal experiences or perspectives (Chung & Slater, 2013; Krakow et al., 2018). ³

Second, narratives that describe the challenges faced by members of a marginalized community may evoke empathy toward the group, increasing positive affect and generating a desire to help by supporting beneficial policies (Adida et al., 2025; Busselle & Bilandzic, 2009; Nabi & Green, 2015). Empathy has been cited as the primary mechanism in a related set of perspective-taking interventions in both survey and field experiments (Adida et al., 2018; Kalla & Broockman, 2023; Williamson et al., 2021). However, some research finds that empathy-generating interventions are most effective among those who already hold positive views of outgroups (Gubler et al., 2022). Treatments designed to increase empathy may wind up “preaching to the choir” and fail to change attitudes among the intended audience (those who hold more negative baseline predispositions) due to cognitive dissonance with the messages (Gubler et al., 2022) or may even provoke backlash (Feldman et al., 2020). Moreover, some research finds that empathy-evoking narratives are more effective at increasing positive affect towards outgroups than shifting policy views, suggesting that empathy alone may be insufficient for overcoming the economic, cultural, or security threat perceptions that drive opposition to refugees and immigrants in some contexts (Bandiera et al., 2026).

Third, personal narratives can provide positive exemplars that lead recipients to update attitudes about the group writ large. Models of exemplar-based attitude formation from psychology suggest that exposure to information about individual group members can shift perceptions of the overall group (Paolini & McIntyre, 2019; Zillmann & Brosius, 2012). One plausible explanation for this is that both stereotypic and counter-stereotypic exemplars often coexist in individuals’ minds. Exposure to information about an individual member can increase the mental accessibility of positive exemplars (Gawronski & Bodenhausen, 2006). Most studies on exemplar-based updating have examined the effects of minimal stimuli, such as short descriptions or images of outgroup members (Dasgupta & Greenwald, 2001; Hamill et al., 1980; Mastro & Tukachinsky, 2011; Ramasubramanian, 2015). We suggest that information conveyed in narrative form could also produce similar shifts.

For these reasons, personal narratives have been posited to be useful for persuasion, including to counter negative attitudes toward immigrant outgroups. Several studies demonstrate that narrative interventions can be effective in combination with other persuasive techniques. Survey experiments have combined personal narratives with facts and statistics (Harney, 2023), news articles and empathy prompts (Wojcieszak & Kim, 2016), and essay writing tasks (Kalla & Broockman, 2023). In a set of notable studies on door-to-door canvassing in the U.S., Kalla and Broockman (2023) and Kalla and Broockman (2020) show that interpersonal conversations including exchanges of personal narratives reduce prejudice towards undocumented immigrants and other marginalized groups. These two-way conversations included several additional components, including eliciting participants’ views, asking them to engage in perspective-taking, meeting them with non-judgmental listening, and correcting misperceptions. This approach was replicated by Larson and Lewis (2025) in Uganda.

This study aims to advance the literature on narrative interventions for countering opposition to refugees and migrants in three specific ways. First, we examine whether exposure to light-touch personal narratives delivered without other intervention elements can be effective. This is a relevant social science question and may also be of interest to advocacy groups seeking to deploy narrative interventions over social and traditional media. We note that recent studies find mixed evidence for narrative treatments in isolation, particularly for interventions delivered without interpersonal contact. Neither Kalla and Broockman (2023) nor Bandiera et al. (2026) find that reading a narrative online shifted policy attitudes towards undocumented immigrants in the U.S. or Colombia, respectively. ⁴ Bandiera et al., 2026 show that an online video treatment is effective, suggesting that mode of delivery may be important, though Gubler et al. (2022) find that an online video emphasizing the plight of undocumented immigrants in the U.S. had limited effects, shifting intergroup attitudes but not policy views. Adida et al. (2025) find positive effects from a narrative about a Somali refugee in the U.S., both with and without the addition of facts and statistics.

Second, this study places particular emphasis on examining whether narrative interventions are effective among those who hold negative predispositions toward the target group. As we discuss in a subsequent section, individuals with positive predispositions could be unaffected if the information contained in the narrative is not new or impressive, and ceiling effects could limit the potential for positive shifts. Individuals with negative predispositions could be unmoved due to motivated reasoning, if the narratives do not work as theorized to circumvent it. Of course, advocacy groups typically wish to counter opposition to refugees and migrants rather than “preach to the choir.” The existing literature has generally not focused on collecting data on predispositions or examining heterogeneous treatment effects. Those that do find mixed results. ⁵ In this study, we collect a set of theoretically-driven measures of baseline dispositions to test whether the narrative interventions shift views among those who view the Somali community as a greater cultural, economic, or security threat and those who have limited contact with the group.

Third, we examine narratives generated by members of the target community and delivered in their own voices. As noted, the treatment scripts were designed to reflect local understanding of the beliefs that underlie negative perceptions and the types of messages thought to be most effective in countering them. The scripts were then recorded as audio treatments by members of each Somali sub-group so that the content could be conveyed by members of the target group. This collaborative approach reflects best practices for scholars working in the Global South (Michelitch, 2018; Thachil, 2018) and has the potential to produce findings that will be useful to local actors. The decision to generate first-person accounts for this study also reflects the cumulative findings from the literature on persuasion, which emphasizes that narrators’ identities can affect perceptions of trust, reliability, and authority—potentially making first-person accounts more persuasive than second-hand accounts provided by other actors (Druckman, 2022; Nünning, 2015; Van Lissa et al., 2016).

Context: Somalis in Kenya

Due to multiple conflicts in nearby countries, Kenya hosts one of the world’s largest refugee populations at roughly a half a million refugees (UNHCR, 2024). Ongoing violence, environmental disaster, and state collapse in neighboring Somalia since the late-1980s has fueled a protracted refugee crisis for over three decades. The largest share of refugees in Kenya (about 55%) are from Somalia. Many live in the Dadaab refugee camp, which at its peak in 2011 was the world’s largest camp with over 500,000 inhabitants and currently holds about 250,000 people. Many also reside in Nairobi’s Eastleigh neighborhood, an area that is synonymous with Somalis.

The Somali community in Kenya also includes more than two million longtime citizens, making up roughly 6% of the Kenyan population. As in other parts of Africa where colonial borders partitioned ethnic communities, borders drawn by European colonial powers divided Somali territory between present-day Somalia, Kenya, and Ethiopia (Khadiagala, 2010). ⁶ These borders remained after independence, engendering irredentist movements seeking to unite the Somali population, including a violent insurgency in Kenya’s Northeastern Province, where Kenyan Somalis predominate (Thompson, 2015). Since the 1960s, the Kenyan government has employed an array of heavy-handed tactics to suppress separatist efforts in the area (Lochery, 2012).

Regardless of citizenship status, Somalis in Kenya are often viewed collectively as an immigrant outgroup—a foreign population that holds an ambiguous place in the Kenyan nation (Lochery, 2012; Otunnu, 1992; Scharrer, 2018). Somalis are subject to discrimination, police brutality, and harassment by state officials (Weitzberg, 2017). Economic tensions have periodically emerged over access to herding territory in northern Kenya and the growing role of the Somali commercial diaspora in Nairobi (Carrier & Lochery, 2013). Differences in religion also distinguish the Somali community: while Somalis are overwhelmingly Muslim, most other Kenyans are Christian.

Since 2010 tensions have been exacerbated by the rise of Al Shabaab, a Somalia-based Islamist insurgent group. Al Shabaab committed an estimated 297 separate attacks in Kenya with a fatality count of 1,025 between 2012 and 2020. ⁷ High-profile attacks include Nairobi’s Westgate Mall (2013), Garissa University College (2015), and Nairobi’s Dusit luxury hotel (2019). Public statements by Kenya’s top political leaders in recent years have linked Somalis with these attacks in the public mind (IRC, 2018). Government officials have repeatedly claimed that Al Shabaab recruits from refugee camps and other Somali areas, portraying these populations as a threat to national security. Citing alleged security concerns, the government first announced its intentions to close Dadaab in 2016, a decision that would have forced hundreds of thousands of Somalis to return to insecure conditions in Somalia.

Media portrayals likely contribute to negative stereotypes that link Somalis with terrorism (IRC, 2018). To examine the predominant frames in public dialogue regarding the Somali community, we conducted content analysis of newspaper articles in Kenya’s leading newspaper, The Daily Nation, for the 12 months prior to our fieldwork (see details in Supplemental Information (SI) B). A keyword search on the stem “Somali-” yielded a sample of 838 articles. Security threat is by far the most frequent frame in public discourse about Somalis in Kenya. As shown in Figure 1(a), an overwhelming majority (76%) of articles referencing the Somali community mentioned violence and insecurity, and of those a large share (82%) mentioned terrorism and terrorist organizations. This pattern holds when we examine articles that mention Somali refugees/immigrants (Figure 1(b)); Somali Kenyans, that is, citizens of Kenya (Figure 1(c)); and Somalis outside Kenya or Somalia the country (Figure 1(d)). Notably, several articles in the sample included quotes from high-level government officials linking Somalis to security threats. For example, one article quoted President Uhuru Kenyatta as saying, “Dadaab has become a protracted situation, characterized by hopelessness…That allows terrorist operatives to exploit it for their operational efforts.” ⁸

Figure 1.

Media portrayal of Somalis in Kenya.

One feature of the Kenyan context makes it especially well suited for exploring the potential of narrative interventions. In contrast to many other contexts, the status of refugees and immigrants has not emerged as a divisive partisan issue in Kenya, and our survey results show little difference in opinion across partisan lines (see SI D.4). Where public attitudes toward refugees and migrants do not overlap with partisan orientations, treatments designed to counter opposition may be more effective than in settings where such views are linked to partisan attachments (Hopkins et al., 2019). At the same time, we think of this as a potentially difficult case for any type of intervention designed to shift attitudes toward Somali citizens and refugees given pervasive and long-standing prejudice toward the group and its vilification as a security threat by public officials and media in the years preceding the study.

Research Design

To examine whether personal narratives offer an effective strategy, we fielded a household survey designed to be broadly representative of Nairobi County, which includes the city of Nairobi and surrounding areas. We selected Nairobi as the research site because it is large (nearly 10% of the national population) and diverse, with most of Kenya’s major ethnic groups represented. Moreover, Nairobi is one of the few locations where Somalis and non-Somalis reside. Outside of Nairobi, ethnic Somalis are concentrated in the Northeast province. It is also the location of major terrorist attacks conducted by Al Shabaab in recent years.

We began by holding a series of informal meetings and interviews with members of the Somali community—both refugees and Kenyan citizens—in Nairobi’s Eastleigh neighborhood (see SI C for details). These conversations were crucial for gaining an understanding of the types of messages Somalis thought might reduce discrimination against them and that were missing from the national discourse. We then collaborated with leaders from Eastleighwood, a local advocacy organization that creates media content designed to bridge divisions between Somali and non-Somali Kenyans, to develop and record audio treatments based on this input.

This effort generated two distinct narratives (recorded in Kiswahili) aimed at addressing distinct concerns articulated by different segments of the Somali community. We made no attempt to hold constant aspects of the two narratives other than length, placing more value on allowing our partners to shape the content. The first narrative—recorded by a recent refugee—reflects the experience of refugees fleeing violence in Somalia in order to evoke empathy by highlighting the reasons for leaving Somalia, the challenges faced in Kenya, and gratitude to the Kenyan nation. The narrative was an attempt to improve positive sentiments toward Somalis and important policies such as keeping the Dadaab refugee camp open and allowing a greater number of refugees to become citizens. The second narrative—recorded by a longtime citizen—reflects concerns about the perception that Somalis in Kenya support groups like Al Shabaab and seeks to dispel such views by communicating shared opposition to terrorism. The first treatment recounts the life experiences of the speaker, while the second conveys the speaker’s personal perspectives. Both types of narratives (biographic versus perspective sharing) have been studied in the communications literature (e.g., Braverman, 2008; Krause & Rucker, 2020). It is important to note that because the treatments vary both the speaker’s identity and the message content, we do not attempt to determine whether they are effective due to the content of the narrative or the identity of the narrator. ⁹

The treatments likely convey information and sentiments that are novel to most non-Somali Kenyans, the target of the intervention. Additional newspaper content analysis found that only 3.3% of articles in the prior year mentioned refugee hardships and only 0.9% mentioned shared opposition to terrorism. The infrequent media coverage of the treatment themes bolsters our expectation that they might plausibly shift public sentiments, since they do not overlap with information or views routinely encountered in mass media. Our survey data also show that non-Somali Kenyans have relatively little direct contact with Somalis: 85% report that they have seldom or never spoken with a Somali refugee in the past year, with 57% having seldom or never had a conversation with a Somali Kenyan in the same time period.

We implemented the experiment as part of a household survey in Nairobi (n = 1,112) on June 6–19, 2017. We used standard sampling procedures to produce a representative sample of the adult population of Nairobi County. ¹⁰ The sampling approach employed random selection of 54 starting points (clusters) allocated across parliamentary constituencies according to population. Within clusters, enumerators followed a random-walk procedure to identify 20 households in which respondents were randomly selected from among those home at the time of the visit. SI D.1 includes details on sampling and shows that the sample closely approximates recent Afrobarometer samples from Nairobi. ¹¹

Respondents were randomly assigned to hear the refugee hardship narrative (T1), the anti-terror narrative (T2), or no narrative (Control). The recordings were conveyed to participants through headphones, administered by enumerators using tablet computers. Because our approach yielded two very different treatments, we created a set of distinct outcome measures tailored to each message. Thus, our goal is not to conduct a head-to-head test that compares the effectiveness of each treatments against the other, but rather to examine whether each treatment shifts views, relative to the control condition. The SI reports descriptive statistics for all variables (SI D.3), basic relationships between outcome attitudes and demographics (SI D.4), and balance on covariates and non-response across treatment groups (SI D.5). ¹²

Treatment 1: Refugee Hardship Narrative

The refugee hardship narrative describes experiences and sentiments common to Somali refugees through the lens of one person’s journey (Figure 2). In a context where refugees are often portrayed in collective terms, the biographical nature of the narrative provides a human face to the refugee crisis and encourages the listener to appreciate the challenges refugees encounter.

Figure 2.

Treatment 1 – refugee hardship narrative.

Outcome measures, listed in Table 1, include questions that probe intergroup affect and policy positions related to several ongoing debates at the time of the study. The measure of intergroup affect is based on a question that asks how close respondents feel to Somalis. Policy measures focus on: (1) government spending on refugees, (2) support for building a border wall, (3) the percentage of refugees that should be granted citizenship, and (4) support for shutting Kenya’s main refugee camp (Dadaab) and repatriating refugees. Of particular interest is whether the narrative has an effect across multiple policy areas, or whether effects are confined to the main issue area discussed in the treatment, closing Dadaab. Outcome variables are rescaled, as necessary, to range from 0 to 1 with higher values indicating more favorable attitudes toward refugees and policies that would benefit them. We also create an index of these items (Cronbach’s alpha = 0.6) based on the average of the non-missing values for the five items. ¹³

Table 1.

Outcome Measures for T1 – Refugee Hardship Narrative.

Closeness	How close do you feel to Somalis? 0 means not very close, 5 means in the middle, and 10 means very close.
Spending too low	In your view, does the government spend too much, too little, or the right amount of money on helping Somali refugees, or haven’t you heard enough to say? Recoded to dichotomous variable: 1 = gov’t spends too little; 0 otherwise.
Oppose wall	How much do you support the current proposed government initiative to build a wall on the border with Somalia to decrease the ability for Somali citizens to enter Kenya? For this, 0 means not at all, 5 means in the middle, and 10 means very much.
Percent citizens	Out of every 10 Somali refugees in Kenya now, how many should be allowed to become Kenyan citizens?
Keep Dadaab	In your view, should the government shut down the Dadaab refugee camp and force the refugees to return to Somalia? [0 = yes, 1 = no]
T1 index	Average of non-missing values for all items.

Figure 3 shows mean values for respondents in the control and the refugee hardship narrative (T1) condition in panel (A) and estimated treatment effects in panel (B). Full regression results are in SI E. Panel (A) shows that attitudes on most outcome measures are relatively unfavorable in the control group: the mean value on closeness to Somalis is 2.94 on the 10 point scale; 45% support keeping Dadaab open; respondents on average support granting citizenship rights to 3 of every 10 Somali refugees; just 20% of control respondents believe that spending on refugee assistance is too low; and only 41% oppose building a wall on the Somali border.

Figure 3.

Effects of T1 – refugee hardship narrative.

Panel (B) shows a positive and statistically significant effect on the composite index and three of the five individual outcomes. Listening to the refugee hardship narrative increased affect toward Somalis, as shown by the results for Closeness, by 5 percentage points. The narrative increased support for keeping Dadaab open rather than repatriating refugees back to Somalia by 17 points. This effect is particularly striking given that, as noted, high-level government officials repeatedly called for its closure in the months before the study. The treatment also increased support for granting citizenship to a larger share of refugees by 7 points. The citizenship status of refugees in Kenya has been an issue of historical importance for several decades, and is of increasing importance given that some refugee families have now been in the camps for multiple generations. The observed effect on citizenship, which is not directly mentioned by the narrative, suggests that the treatment may influence views on related issue areas.

The treatment has no significant effect on two policy issues: perceptions of government spending and the border wall. Like citizenship, these policies are not specifically mentioned in the treatment, suggesting that the ability to influence views on related issues may be limited for outcomes that are less directly related to its content. Yet, it is also possible that the treatment did not affect these attitudes for more nuanced reasons related to the policy issues at hand. The null result for government spending may be due to the fact that the bulk of refugee funding comes from the international community rather than the government, or to the common sentiment that Kenya does more than its fair share in hosting refugees compared to other countries (IRC, 2018). Alternatively, views on government spending could be unaffected because the speaker indicates that while he faces many challenges, “life in Dadaab was much better than back in Somalia.” A narrative emphasizing persistent hardships in the camp and the need for greater funding might evoke a stronger reaction among listeners on this outcome. The null effects on the border wall may reflect the view that keeping the border open would offer no positive benefit for current refugees already in Kenya—those featured in the treatment. ¹⁴

Treatment 2: Anti-Terror Narrative

The anti-terror narrative describes the speaker’s opposition to Islamist groups like Al Shabaab, seeking to dispel the notion that Somalis sympathize with the group’s objectives and methods (Figure 4). The message emphasizes a superordinate identity, stressing that opposition to violent extremism is common across religious faith communities in Kenya. Prior research suggests that priming superordinate identities can reduce intergroup bias, including in African contexts where national identity is often weak (Gaertner et al., 2000; Robinson, 2016). The script frames the “bridging” message in terms of religion, rather than ethnicity. While Somalis are distinct with regard to both identity dimensions, the links to terrorism are often understood in religious, not ethnic, terms. Finally, as noted above, the anti-terror narrative is less biographic in nature than the refugee hardship narrative, conveying the speaker’s personal perspectives rather than his life history. Unlike T1, it does not highlight the suffering of Somalis or seek to evoke empathy.

Figure 4.

Treatment 2 – anti-terror narrative.

Outcome measures, shown in Table 2, again include a range of intergroup and policy attitudes designed to reflect the specific content of the anti-terror narrative and long-standing questions about the inclusion of the Somali community in the Kenyan nation. Because the anti-terror narrative was designed to combat beliefs that Somalis support terrorism, we employ a larger set of perception measures than for the refugee hardship narrative. These include: (1) group affect measured by closeness, (2) beliefs about Somalis as violent or peaceful, (3) beliefs about Somalis as nationalist versus tribalist, and (4) perceptions of support among Somalis for terrorist groups. Policy-oriented questions probe views about: (5) whether Somali Kenyans should be forced to leave Kenya, and (6) whether Kenya would be better off without Northeastern Province, the predominantly Somali border region. As noted earlier, debates about the place of Somalis in Kenya and the status of Northeastern Province have been sources of recurring tension since before independence in 1963. These policy issues are not directly addressed in the narrative, though they are related to basic questions about inter-communal relations between Somalis and others in Kenya. We rescale all items to range from 0 to 1, with higher values indicating more positive attitudes, and create a composite index from the items (Cronbach’s alpha = 0.6) that averages non-missing values. ¹⁵

Table 2.

Outcome Measures – T2: Anti-Terror Narrative.

Closeness	How close do you feel to Somalis? 0 means not very close, 5 means in the middle, and 10 means very close.
Peaceful	Would you describe Somalis as peaceful or violent? For this, 0 means very peaceful, 5 means in the middle, and 10 means very violent.
Oppose terrorism	In your opinion, how many Somalis in Nairobi support Islamic extremists like Al Shabaab? [0 = all, 1 = many, 2 = some, 3 = a few, 4 = none]
Nationalist	Would you describe Somalis as nationalist, caring about all Kenyans, or tribalist, caring only about their own tribe? For this, 0 means very nationalist, 5 means in the middle, and 10 means very tribalist.
Remain in Kenya	In your view, should Kenyan Somalis leave Kenya and go to Somalia? [0 = yes; 1 = no]
Better w/Northeastern	Do you think Kenya would be better off without the Northeastern Province? [0 = yes; 1 = no]
T2 index	Average of non-missing values for all items.

Figure 5 shows mean values for the control and the anti-terror narrative (T2) groups and estimated treatment effects, with full regression results in SI E. We again observe relatively low values among control respondents. Mean values on Closeness, Peaceful, and Nationalist are 2.94, 3.76, and 3.27 respectively, on 10-point scales where higher values indicate more positive views. Respondents in the control condition on average believe that “some” Somalis support terrorism (on a scale that ranges from none to all). On the two policy measures, we observe only moderate support for the view that Somalis should remain in Kenya: although a majority of respondents in the control condition believe that Somalis should remain, more than a third (42%) support their expulsion. Of equal concern, a follow-up question reveals that among those in the control condition who expressed support for Somalis leaving Kenya, a substantial share (39%) registered support for forcing Somalis to leave even if they chose not to do so voluntarily. The data show strong support for the idea that Kenya is better off with Northeastern Province, likely because even those who might favor expelling Somalis would not support territorial concessions.

Figure 5.

Effects of T2 – anti-terror narrative.

Estimated treatment effects in Figure 5(B) show that the anti-terror narrative had positive effects across all outcome measures and the composite index, significant at the p < .05 level or lower for all items with the exception of Better w/Northeastern, which is significant at p < .10. The treatment improved general affect as measured by Closeness by 9 percentage points, and altered scores on other perceptions measures, increasing views of Somalis as peaceful and nationalists by 10 and 9 points, respectively. The treatment also produced a small positive effect (4 points) on beliefs about the share of Somalis that oppose terrorism. Given the low to moderate values on these indicators in the control group and the routine linking of Somalis and terrorism by the Kenyan media, even small shifts in absolute terms represent substantively meaningful changes.

We also observe positive effects on the two policy outcomes. The treatment increased the view that Somali Kenyans should remain in Kenya by 14 points. This finding is particularly compelling, indicating that personal communication designed to dispel notions that Somalis pose a security threat can meaningfully counter views that the community does not belong in Kenya. The anti-terror narrative also increased the share of respondents reporting that Kenya is better off with Northeastern Province as part of the county by 6 points—a striking finding given the high mean value in the control and the possibility that ceiling effects might limit the potential for change on this outcome.

Do Personal Narratives Affect the Negatively Predisposed?

Do the personal narratives affect only those favorably predisposed toward Somalis, or do they also affect those who hold less positive views? Exploring heterogeneity is important for both theoretical and practical reasons. Theoretically, we are interested in understanding whether prior dispositions constrain attitudinal shifts. The literature suggests competing ex ante expectations for whether and how prior dispositions should matter in the context of our treatments (Adida et al., 2018; Grigorieff et al., 2020; Gubler et al., 2022; Hopkins et al., 2019; Williamson, 2020). Those who are favorably predisposed may be more receptive to the narratives, resulting in positive effects. However, if their opinions are already at or close to a positive “ceiling,” then there will be less room to grow. In this case, the narratives may simply reinforce existing attitudes. Among the negatively predisposed, individuals have more room for positive attitudinal change; however, they may be more inclined to reject or dismiss the information due to motivated reasoning, producing more modest or null effects.

Beyond these theoretical questions, the exploration of conditional effects also has practical importance. Organizations that advocate on behalf of immigrant outgroups often target those with negative predispositions in order to raise awareness of the plight refugees and immigrants face, and to increase public support for policies that aid them. Moreover, in the specific context of our study, the Somali respondents we interviewed in Nairobi expressed an explicit desire to counter negative views. It is important therefore to test whether personal narratives can alter perceptions among those who hold less positive attitudes.

We employ four potential moderators—listed in Table 3—that reflect the predominant explanations for opposition to refugees and immigrants in the literature (Dancygier & Laitin, 2014; Hainmueller & Hopkins, 2014). Security threat is measured with an implicit association test (IAT) conducted with respondents prior to the main survey. The IAT has been widely used in psychology and political science research to assess implicit associations, including for immigrant outgroups (Pérez, 2010). We adapted the IAT to the Kenyan context in ways described in SI G.1. Our implementation measures implicit views of Somalis as “peaceful” versus “violent.” We opt for an implicit measure because the outcomes used for the anti-terror narrative include two explicit questions that probe beliefs about whether Somalis support violent extremism. Labor market threat is measured with a question that examines beliefs about whether Somalis take jobs from Kenyan citizens in Nairobi. ¹⁶ Cultural threat is measured with a question that probes beliefs about religious dissimilarity between Muslims and Christians. Contact is measured with two questions that ask respondents how often they have interacted with “Somali refugees” and “Somali Kenyans” in the past year; we employ these measures as moderators for each treatment respectively. The four moderators are only weakly correlated, suggesting that they are not merely reflections of a single underlying orientation or trait (see SI G.2). ¹⁷

Table 3.

Moderating Variables.

Security threat	Implicit Association Test (IAT) with “peaceful” and “violent” as target words.
Labor market threat	In your view, how much are Somali refugees reducing the employment prospects of Kenyan citizens here in Nairobi? For this, 0 means not at all, 5 means in the middle, and 10 means very much.
Cultural threat	From what you know, do you think that the Muslim religion and the Christian religion have a lot in common, some things in common, a little in common, or nothing in common?
Contact	How often did you have a conversation with a [Somali refugee/Kenyan Somali] in the last year?

We dichotomize each potential moderator into low (below the median for each item) and high (median and above) values. We present results from models that use the composite indices for each treatment as dependent variables and interact each treatment indicator with the moderating variables sequentially. Regression results are provided in SI G.4. We also provide results from additional tests in the SI that: (1) employ continuous versions of the moderators (SI G.5), (2) present results on individual outcomes for each treatment (SI G.6), and (3) examine whether the conditional effects are confounded by differential social desirability perceptions (SI G.7). We also present results for additional demographic moderators listed in the pre-analysis plan in SI G.8.

The results in Figure 6 indicate that the effects are not confined to respondents who hold more positive attitudes toward the Somali community. The effects of each treatment are in most instances as large among respondents who perceive Somalis to be a greater security, economic or cultural threat and who have less contact with the group. Indeed, for the refugee hardship narrative, positive effects on the composite index are significant only for respondents who are less favorably predisposed on most moderators. The one exception is security threat, where we find positive and significant effects only among those who perceive Somalis as less threatening. Given that associations with terrorism are likely the principal source of opposition to Somali refugees for many Kenyans, this finding is noteworthy. However, examining the individual outcome measures (SI G.6), we observe positive and significant effects among those who associate Somalis with violence on support for keeping Dadaab open (the main policy outcome discussed in the refugee hardship narrative), though the effects are smaller than among those who associate Somalis with peace (0.12 vs. 0.19; difference = 0.07, p < .05). Nonetheless, these findings suggest that there may be limits to narrative interventions that do not directly address the primary source of opposition to refugees and migrants, in this case security concerns. For the anti-terror narrative, positive effects are observed both among those more and less favorably predisposed toward Somalis. It is noteworthy that the anti-terror narrative proved effective even among those on the higher end of the security threat measure, unlike the refugee hardship narrative. This finding reinforces the conclusion that narratives that directly address the main source of opposition may prove more effective than those that stress hardships faced by members of the target group.

Figure 6.

Treatment effects conditional on moderators.

Robustness Tests

We conduct a series of robustness tests to allay concerns about two common threats to causal inference in survey experiments (see SI F for more extensive discussion and results). ¹⁸ First, it is possible that the treatments activate social desirability biases, making respondents less willing to express negative views of the Somali community, rather than altering attitudes. We doubt this explanation for the simple reason that respondents—including those in treatment groups—do not appear to be inhibited in expressing negative beliefs about Somalis or voicing support for policies that would negatively affect Somalis, including forced expulsion.

Nonetheless, to address concerns about this type of bias, we conduct several tests. First, we show that the treatment effects are robust to dropping interviews in which bystanders were present (37% of the sample) and that the treatment effects are not systematically conditioned by bystander presence (SI F.1). If we found that outcome responses were consistently correlated with the presence of bystanders or only found treatment effects (or larger effects) among interviews with bystanders, we might suspect that social desirability bias is driving the results. Second, we speculate that respondents might be less comfortable expressing negative sentiments toward any outgroup in the presence of a non-co-ethnic interviewer. We find, however, that enumerator ethnicity is unrelated to the outcome variables and does not systematically condition the treatment effects (SI F.2). Third, we use a battery of questions that measure respondents’ beliefs about whether neighbors would disapprove if one expressed negative views of various ethnic groups, political parties, and prominent figures. ¹⁹ From these we construct an index that arrays respondents on a continuum in terms of their level of concern about social sanction if they were to make potentially offensive remarks about outgroups. We find that treatment effects are no different among those with high versus low perceptions of social desirability pressures (SI F.3).

In addition, we examine the effects of each treatment on outcome variables tailored for the other narrative. If our results are driven merely by social desirability biases, we should observe that any positive message about Somalis should generate consistent shifts across outcome measures, even if the contents are not related to the outcomes. Results in SI F.4 show that the effects are more limited when we switch outcome variables in this way. Specifically, we observe that the anti-terror narrative positively affects several outcomes crafted for the refugee hardship narrative, which is not surprising since attitudes toward refugees are linked to concerns about terrorism. However, the refugee hardship narrative has no consistent effect on anti-terror narrative outcomes, likely because the content of this narrative is unrelated to the anti-terror narrative’s outcome variables. Collectively, these results mitigate concerns that the findings are driven by social desirability bias.

A second potential concern relates to interviewer demand effects. We reason that if respondents who were exposed to the narratives felt compelled to report positive attitudes toward Somalis in order to please enumerators, we would observe consistently positive responses among the treated. However, on most outcome items, mean responses among treated participants were at the lower end of the scales—they were simply less negative than in the control. Further, if people felt compelled to give the “correct answer,” one would expect to see that the treatments shift all attitudinal outcomes relative to the control. Yet, we observe variation across outcomes, with more consistent effects on outcomes more directly related to the treatments. Finally, we note that all respondents answered a large battery of common questions about Somalis prior to the survey experiment, suggesting that the narratives likely did not differentially alert those in treatment conditions to the nature of the study.

Discussion and Conclusion

This study investigates whether first-person personal narratives are effective in combating negative attitudes toward immigrant outgroups. In collaboration with a local advocacy group in Nairobi, Kenya, we developed two audio treatments that reflect the experiences and views of two Somali sub-groups, recent refugees that have fled violence in Somalia and long-term Kenyan citizens of Somali descent. Results from an experiment embedded in a large household survey show that the narratives produced statistically and substantively significant positive effects on a range of intergroup and policy attitudes among non-Somali Kenyans. Notably, the interventions generally worked equally well among respondents who were more negatively predisposed against the Somali—those who viewed Somalis more negatively on measures of economic, cultural, or security threat, and who reported less frequent contact with Somalis.

These results relate to a growing body of scholarship on narrative strategies for combating opposition to refugees and immigrants. Many of the most successful interventions aimed at reducing negative attitudes bundle narratives with other persuasive techniques (e.g., interpersonal exchanges and perspective-taking exercises) in the context of door-to-door campaigns or school programs (Alan et al., 2021; Kalla & Broockman, 2020; Larson & Lewis, 2025). This study offers further support for narratives in bundled interventions and also suggests that light-touch interventions implemented without those other features can be effective at nudging attitudes—even among the negatively predisposed. Given that light-touch narratives are affordable and scalable, a priority for future research should be to confirm the promise of this strategy by testing narratives delivered through media channels in more naturalistic field settings.

Further research is needed to identify the conditions under which light-touch narratives will be most effective. The findings from this study and related research suggest that three factors are likely to be important: narrative content, context, and delivery mode. First, with regard to narrative content, we speculate that one explanation for the positive results reported in this study may be the local relevance of the narrative, which was developed and recorded by individuals from the target group. The importance of narrative content is illustrated by the varying findings in other work. For example, in survey experiments conducted in the U.S., Adida et al. (2025) find that a narrative from a Somali refugee produced positive effects, while Kalla and Broockman (2023) find that a narrative about an undocumented Latin American immigrant yielded null results. ²⁰ Future research should explore how varying narrative content, including narrator identity, affects outcomes. ²¹

Second, narrative interventions may be more effective in contexts where views are more malleable. In Kenya, our positive results may be due to the fact that there is little overlap between partisan orientations and views on immigration. However, in the Global North, immigration views are often polarized on partisan lines, and for this reason attitudes may be less responsive to light-touch narrative interventions. In such contexts, bundled, high-intensity interventions may be necessary (Bruneau & Saxe, 2012; Kalla & Broockman, 2020, 2023).

Finally, the delivery mode may also explain the mixed results in prior survey-experimental literature. Communications research suggests that narratives delivered as audio or video treatments tend to be more effective than written versions (e.g., Bilandzic & Busselle, 2013; Braverman, 2008). Consistent with this expectation, Bandiera et al., 2026 find that in Colombia the same script produced larger and more consistent shifts when presented as a video rather than written text. Our intervention was delivered as part of a household survey in which respondents were asked to listen to an audio treatment using headphones—an approach that likely increased respondent attentiveness. A promising area for future research would be to examine how delivery mode affects intervention results across a range of contexts where different media formats (radio, television, print media, social media, etc.) are more or less prominent.

While the results from this study are encouraging, several questions remain. First and foremost, additional research is needed to understand the mechanisms through which personal narratives affect respondents’ attitudes toward immigrant outgroups. While personal narratives have been studied extensively in other fields, we still have too little understanding of the cognitive and emotive channels through which narratives from marginalized communities change perceptions. Much of the literature on narratives and related interventions focuses on empathy as the primary mechanism through which change occurs. However, our results for the anti-terror narrative (treatment 2) show that narratives that are not designed to evoke empathy can also be effective, suggesting that other mechanisms are worth exploring—such as emphasizing shared values or directly addressing perceived threats. Second, questions remain about the duration of observed effects. To date, only a few studies have examined whether effects persist over time (Facchini et al., 2022; Grigorieff et al., 2020; Kalla & Broockman, 2020; Larson & Lewis, 2025). Finally, research should continue to examine the value of combining narratives with factual information designed to combat misperceptions about immigrant outgroups, building on Adida et al. (2025). While much remains to be done, the findings described here bode well for those working to combat opposition to refugees and migrants.

Supplemental Material

Supplemental Material - Personal Narratives Reduce Negative Attitudes Toward Immigrant Outgroups: Evidence From Kenya

Supplemental Material for Personal Narratives Reduce Negative Attitudes Toward Immigrant Outgroups: Evidence From Kenya by Nicole Audette, Jeremy Horowitz and Kristin Michelitch in Comparative Political Studies