Environmental Literature as Persuasion: An Experimental Test of the Effects of Reading Climate Fiction

ABSTRACT Literary works of fiction about climate change are becoming more common and more popular among critics and readers. While much research has indicated the persuasive effectiveness of narrative storytelling in general, empirical research has not yet tested the effects of reading climate fiction. This paper reports results from the first experimental study to test the immediate and delayed impacts of climate fiction on readers’ beliefs and attitudes about climate change. We found that reading climate fiction had small but significant positive effects on several important beliefs and attitudes about global warming – observed immediately after participants read the stories. However, these effects diminished to statistical nonsignificance after a one-month interval. In this paper, we review the relevant literature, present the design and results of this experiment, and discuss implications for future research and practice.


Introduction
Increasingly, scholars of environmental communication are calling for creative forms of climate communication (e.g. Boykoff, 2019), including films, television shows, theater, art, and literature. Nearly half of Americans report reading a novel or short story at least once a year (National Endowment for the Arts, 2015), but environmental literature has received relatively little attention in empirical environmental communication research. This research gap is not due to lack of content for analysis, as there are now hundreds of widely-read works of "climate fiction" (Johns-Putra, 2016). While many of them are written with an activist desire to raise awareness and generate behavioral and political change (Schneider-Mayerson, 2017), the effects of reading climate fiction have not yet been studied in controlled experiments. Given the sudden pervasiveness and potential persuasiveness of climate fiction, together with the growing urgency of climate change, empirical research on this subject is needed.
fiction is not newauthors have been imagining climate futures for almost as long as climate change has been a topic of public concern (Trexler, 2015). But in the 1990s, the number of novels that featured anthropogenic climate change as a central element began to grow. By the late 2000s, as climate change impacts became more evident and public concern intensified, there were dozens of Englishlanguage novels published (Schneider-Mayerson, 2017;Trexler, 2015). Today, the genre is growing so quickly that any comprehensive list would soon become outdated. These works are no longer relegated to the margins of literature or viewed as mere genre fiction, but instead represent some of the most well-respected and widely-read English-language authors writing today, including Margaret Atwood, Barbara Kingsolver, Amitav Ghosh, Ian McEwan, Cormac McCarthy, David Mitchell, and Kim Stanley Robinson. Fictional works that focus on climate change are also beginning to win major literary awards. For example, Lauren Groff's book of short stories, Florida (2018), was a finalist for the National Book Award for fiction, and Richard Powers was awarded the Pulitzer Prize for The Overstory (2018).
Scholars of environmental communication and environmental psychology have argued that the arts can play a critical role in influencing beliefs, attitudes, and behaviors related to climate change (e.g. Boykoff, 2019;Gabrys & Yusoff, 2012;Milkoreit, 2016). For example, Corbett and Clark argue that the arts allow "the so-called invisibility of climate change to be seen, felt, and imagined in the present and the future" and "encourage critical reflection on existing social structures and cultural and moral norms" (Corbett & Clark, 2017). Many critical scholars of environmental literature have speculated about the potential impact of climate fiction on individual readers (e.g. Gaard, 2014;Goodbody, 2014;Mehnert, 2016), and this consideration is often implicit in the growing attention to climate fiction in literary criticism (Johns-Putra, 2016;Schneider-Mayerson et al., 2020).
However, the extant empirical scholarship on the influence of environmental art on its audiences has primarily focused on film (e.g. Howell, 2011;Leiserowitz, 2004) and fine art (e.g. Curtis, 2009;Sommer & Klöckner, 2019). Accordingly, environmental communication researchers have called for more study of the effects of communicating through the environmental arts (Moser, 2016), suggesting that stories might engage people in novel and persuasive ways.

The persuasive power of narrative
Narrative persuasion entails embedding a persuasive message within a story that has both a plot and identifiable characters . A growing body of research indicates that narrative story-telling is an effective mode of communicating persuasive messages, particularly in science communication contexts (Dahlstrom, 2014;Shen et al., 2015). Scholars theorize two central mechanisms by which a narrative format enhances the persuasive effects of a message: identification and transportation (Cohen et al., 2015;. Identification refers to the connection the audience feels to the characters in a narrative. Several studies have operationalized this construct (and analogous constructs) with measures that include elements such as perceived similarity between the reader and character, the reader's understanding of the character, and their shared emotional experience (e.g. Busselle & Bilandzic, 2009;Cohen, 2001;De Graaf et al., 2012;Moyer-Gusé, 2008;Tal-Or & Cohen, 2010). Because identification with a narrative's character(s) is motivated by perceived similarity, it is likely that the alignment of worldviews, values, or ideology between the audience and the character can impact the level of identification that the audience feels. For narratives presenting a message about riskas is the case with many stories about climate changeevidence suggests that identification increases risk perceptions by decreasing perceived social and psychological distance between the audience and the narrative's characters (So & Nabi, 2013).
Narrative persuasion is also considered to be effective due to increased transportation, which is defined by Green and Brock (2000) as a state in which "all mental systems and capacities become focused on events occurring in the narrative" (p. 701). A meta-analysis of 76 studies (Van Laer et al., 2013) found that transportation from narratives has large effects on emotions, and moderate effects on beliefs, attitudes, and behavioral intentions (e.g. Appel & Richter, 2010;Green, 2004;Green & Brock, 2000;Vaughn et al., 2009). Scholars argue that transportation can enhance persuasion by reducing counterarguing (conscious consideration of rebuttals to the message) andas with identificationby decreasing psychological distance between the audience and the content described in the narrative (Moyer-Gusé, 2008;So & Nabi, 2013).
While research has found narratives to be persuasive in pro-social, non-controversial messages such as public health campaigns (e.g. Murphy et al., 2013;Niederdeppe et al., 2011), several other experiments have also found narratives to be persuasive in polarizing, contentious topics such as immigration, stem cell research, gay rights, affirmative action, and the Israeli-Palestinian conflict (Boswell et al., 2011;Cohen et al., 2015;Mazzocco et al., 2010;Zhang & Min, 2013). Many of these studies specifically tested the explanatory mechanisms of transportation and identification. For example, Igartua and Barrios (2012) found that identification mediates the relationship between narrative appeals and attitude change about one's own religion.
In addition to being effective across myriad topics, narrative persuasion also appears to be effective across multiple types of narratives. Studies indicate that narratives presented as being fictional are no more or less effective than narratives presented as nonfiction (Appel & Malečkar, 2012;Strange & Leung, 1999;Wheeler et al., 1999). Together, the large body of research on narrative persuasion supports the speculations of environmental scholars that climate fiction stories could have significant persuasive effects on readers' beliefs, attitudes, and behaviors.

The understudied influence of climate change fiction
Only a few studies have investigated the effects of narratives and other literature in environmental topics. For example, studies indicate that reading environmental narratives about animals can cause a boost in pro-animal attitudes in student samples (e.g. Małecki et al., 2018;Małecki et al., 2019). Further, Cooper and Nisbet (2016) report that narratives told through news stories, documentaries, and fictional television shows about GMOs and fracking resulted in increased risk perceptions and support for environmental policy via increased narrative involvement and negative affect. Similarly, Appel and Mara (2013) found that narratives about sustainable driving behavior increased intentions toward fuel-efficient driving, particularly when the narrative's characters were portrayed (and perceived) as trustworthy. However, another study (Jones, 2014) found that narratives about climate change policy were no more persuasive than a list of facts.
Despite the growing corpus of climate fiction and the evidence indicating the persuasive power of communicating about the environment through narratives, only two studies on the influence of these literary works of climate change fiction have been published. Using an exploratory qualitative survey, Schneider-Mayerson (2018) found that climate fiction readers reported that reading climate fiction caused them to perceive climate change as more psychologically proximate, to explore the psychological and social dimensions of climate change, and to appreciate the scale and gravity of climate change. A second survey (Schneider-Mayerson, 2020) suggests that a climate fiction novel can make readers more aware of climate injustice and lead them to empathize with climate migrants. However, both studies were surveys that asked readers' to describe the ways in which they were (or were not) affected, rather than to directly measure the effects of reading climate fiction in a controlled experiment. In the present study, we address this gap in the literature with a controlled experiment testing the influence of reading climate change fiction on a national sample of fiction readers.
Given the extant evidence of the effectiveness of narratives in persuasion, we expect to find that (H1) participants who read climate fiction stories will report more pro-climate beliefs and attitudes, relative to participants in a control group. Following from the empirical and theoretical research on the mechanisms of narrative persuasion, we expect that (H2) the attitudinal and behavioral effects of reading climate fiction will be mediated by identification and transportation.
Most persuasion research assesses the immediate effects of a treatment message, and thus ignores the question of the longevity of those effects. Some tests of narrative persuasion have found long-lasting effects on audiencessuch as when the attitudes were measured a few weeks after the narrative (Appel & Mara, 2013;Appel & Richter, 2007;Jensen et al., 2011)while others have found that effects last for a week, but disappear within a month (Małecki et al., 2019). Because this research is sparse and understanding of the longevity and stability of message effects is still in its infancy, we do not offer hypotheses but instead ask (RQ1): After one month, how do the climate change beliefs and attitudes of participants who read climate fiction stories compare to those who did not?

Population of interest
The population of interest for this experiment was fiction readers who tend to believe that global warming is happening and human-caused, but do not consider the risks to be severe and do not consider the issue to be highly important to them personally. These are characteristics of the "Concerned" and "Cautious" segments of Global Warming's Six Americas (Maibach et al., 2009). This focus heightens ecological validity. That is, we were not interested in measuring the effect of reading climate fiction on those who tend to doubt or dismiss the reality of anthropogenic climate change (e.g. the "Doubtful" or "Dismissive" segments), or on those who do not have an interest in reading fiction, because such individuals are less likely to read climate fiction outside the current experimental context (Schneider-Mayerson, 2018). We also did not test the effect of climate fiction on individuals who already have high levels of worry and perceived risk regarding climate change (the "Alarmed" segment of the Six Americas), because they tend to be highly concerned about the issue already.

Design and procedure
This study involved an experiment designed to test the effect of reading climate fiction on beliefs and attitudes about global warming. The experiment was constructed with two treatment groups and one control group for between-groups comparisons, and using within-subjects measurements that spanned three time points (pre-post-post). First, a Time 1 (T1) screener survey identified participants who are fiction readers and belong to the Cautious or Concerned segments. Those who satisfied these criteria were eligible to participate in the Time 2 (T2) main study. About two weeks after T1, eligible participants were invited into the T2 portionwhich contained a pre-test, the treatment stimuli (reading a climate fiction short story), and an immediate post-test. About one month later, participants who completed T2 were invited to participate in a delayed post-test (T3) to assess the longevity of the effects of reading a climate fiction story at T2.
At T2, participants first completed a battery of pre-test measures of beliefs and attitudes about global warming. Then participants were randomly assigned to one of three experimental conditions. In two of these conditions, participants were assigned to read one of two climate fiction short stories. In the third condition (the control), participants read a fictional short story that was not about climate change. After reading their assigned story, participants were given the same measures of beliefs and attitudes, some additional measures of their experience with and perceptions of the fiction stories themselves, and, finally, several demographic questions. Participation in the delayed T3 involved completing, for a third time, the battery of measures of beliefs and attitudes about global warming.
Before participating in each time point, participants gave informed consent after being made aware of the confidentiality of their response data, the protection of their anonymity, and the optional nature of their participation. All participants were compensated immediately after each portion of the study that they completed (compensation: T1 = $0.20; T2 = $2.50; T3 = $0.50).

Sample
Participants were recruited from Amazon's Mechanical Turk (MTurk), a pool of workers who perform small tasks (surveys, data entry, etc.) for small monetary compensation. Samples recruited from opt-in online panels like MTurk are not nationally representative, but they are more diverse than student samples and perform similarly to nationally representative samples on many tasks (Bartneck et al., 2015;Hauser & Schwarz, 2016;Kees et al., 2017). MTurk also affords the ability to select participants who gave particular answers on a prior survey and recontact them with an invitation to participate in another survey, which was a necessary functionality for the longitudinal design of this study.
When recruiting participants for T1, we used the TurkPrime interface to automatically exclude all MTurk workers who had previously completed any of our several prior surveys and experiments about climate change, to exclude all IP addresses flagged by MTurk as suspicious, and to exclude all MTurk workers located outside of the United States. In total, 5,655 American adults entered the T1 survey. Of these, 18 cases were removed for incomplete responses, and 78 cases were removed due to 39 instances of duplicate IP addresses. This left 5,561 valid cases for analysis. Of these, 1,984 (35.7%) satisfied the inclusion criteria for Time 2 (T2): they were Concerned or Cautious according to the Six Americas segmentation, they indicated that they read "occasionally" and were at least "moderately" interested in reading fiction, and they indicated that they were willing to participate in T2.
After about a two-week delay, these 1,984 qualified T1 participants were invited to participate in T2. The re-participation rate was 84% (N T2 = 1,671). To reduce the potential for selective participation biases and demand effects, the T2 recruitment posting on MTurk targeted T1 participants without revealing to them that T2 was related to T1. Of the participants who entered T2, 1,507 (90%) completed the survey. After data cleaning (see Supplementary Material for details) 1,294 valid T2 cases for analysis (n = 434 in Control, n = 469 in "Flight" treatment, n = 391 in "Tamarisk" treatment). This final sample was predominantly (77%) white, (65%) female, and skewed younger and more liberal than the American population (full sample demographics are provided in the Supplementary Material). The skew of the sample toward liberal and female individuals reflects the population of interest, as Americans in the Concerned and Cautious segments tend to be more liberal than the national population, and Americans who read literature at least once per year are more likely to be (62%) female (NEA, 2015).
Approximately one month later, these 1,284 participants who completed T2 were invited to participate in T3. Similar to the T2 recruitment, the T3 recruitment posting on MTurk did not mention that the task was connected to the prior T2 or T1 tasks. Of the 1,284 valid T2 cases, 744 (57%) completed T3 after the one-month delay (n = 300 in Control, n = 299 in Flight, n = 269 in Tamarisk). This attrition did not substantially alter the demographic or ideological sample characteristics (see Supplementary Material for a detailed comparison).

Stimulus materials
Participants in one treatment condition ("Flight") read Helen Simpson's "In-Flight Entertainment" (Simpson, 2011) and participants in the other treatment condition ("Tamarisk") read Paolo Bacigalupi's "The Tamarisk Hunter" (Bacigalupi, 2008). We chose to test two very different short stories (instead of just one story) so as to enable greater generalizability of observed effects. These short stories fit into different categories of Schneider-Mayerson's (2017) typology of American climate fiction. "In-Flight Entertainment" (word count: 4,418; reading time: mean = 18:07; median = 17:27) is a realist narrative that explores the psychological and moral dimensions of the "denial, avoidance, and acceptance" of the gravity of climate change in the present (Schneider-Mayerson, 2017). The story does so by staging a conversation about climate change between a skeptical wealthy businessman and a former climate scientist in the First Class section of an intercontinental airplane, while they complain about the inconvenience of an elderly passenger who is having a heart attack nearby. By dramatizing the lack of appropriate urgency about and response to climate changeliterally and metaphoricallythe story indirectly implicates the reader. In contrast, "The Tamarisk Hunter" (word count: 5,095; reading time: mean = 18:10; median = 17:05) is a prototypical "cautionary fable of the Anthropocene" (Schneider-Mayerson, 2017), placing the reader in a dystopic, climatechanged future. The speculative fiction story is set decades in the future in an American Southwest characterized by climate apartheid, where cities are drying up and disappearing and states are fighting for scarce water and have closed their borders to migrants from other states. These works were selected for their engaging storytelling and their ecological validity -Bacigalupi and Simpson are two widely-read and respected authors of environmentally-engaged fictionand because they represent different modes of climate fiction storytelling. This allows us to observe whether different kinds of climate fiction narratives might have significantly different persuasive effects; or whether there is a consistent pattern of effects. Both stories were lightly edited to enhance clarity, and "In-Flight Entertainment" was edited to make its protagonists American, since perceived similarity has been shown to affect character identification (Brown, 2015).
Participants in the control condition read the short story "Good People" by David Foster Wallace (2007). This story (word count: 3,211; reading time: mean = 13:53; median = 12:46) portrays a young couple in a tense scene of introspection and conversation, as they both ponder their futures and the available courses of action in light of their recent pregnancy and their plan to have an abortion. This story was similar in length to the treatments, was also written by an established author of short stories, but did not touch on environmental or climate issues. The full texts of all three stories are presented in the Supplementary Material.

Global warming beliefs and attitudes
At the beginning of the T2 pre-test, participants were first shown a definition of global warming (see the Supplementary Material for the full text). Then we administered self-report measures of beliefs and attitudes about global warming, which were adapted from those used in the biannual Climate Change in the American Mind survey (e.g. Leiserowitz et al., 2019). The full text of each question and all response options is presented in the Supplementary Material. We measured belief in the existence of global warming using a seven-point scale with anchors of "I strongly believe global warming is NOT happening" (1), "I am unsure whether or not global warming is happening" (4), and "I strongly believe global warming IS happening" (7). Next, we measured belief in the anthropogenic nature of global warming using a seven-point scale with anchors of "I believe global warming is caused entirely by natural changes in the environment" (1), "I believe global warming is caused equally by natural changes and human activities" (4), and "I believe global warming is caused entirely by human activities" (7).
Next, we measured worry about global warming, two types of risk perceptions, and personal importance. These four items produce the Six Americas segmentation (Chryst et al., 2018). Participants indicated their worry about global warming on a four-point scale from "Not at all worried" (1) to "Very worried" (4), and indicated how much they think global warming will harm themselves and future generations, respectively, on separate four-point scales ranging from "Not at all" (1) to "A great deal" (4), with an additional "Don't know" option. We measured personal importance of global warming by asking "How important is the issue of global warming to you personally?" with responses given on a five-point scale ranging from "Not at all important" (1) to "Extremely important" (5).
Because both of the climate fiction stimuli mentioned or portrayed the future effects of global warming, we also measured perceived severity of these future effects by asking how global warming the frequency of "Droughts and water shortages," "People living in poverty," "Refugees," and "Floods" over the next 40 years. Responses to these items were given on five-point scales with response options ranging from "Many less" (1) to "Many more" (5).
Research shows that persuasive messages about threats like climate change should not just focus on the threat, but also promote self-efficacy (perception that one is capable of performing the recommended solution) and response efficacy (perception that the recommended solution will be effective; e.g. Nabi et al., 2018;Witte & Allen, 2000). Such efficacy perceptions are predictors of support for climate change solutions (e.g. Roser-Renouf et al., 2014). Therefore, we measured efficacy perceptions by asking, "In your opinion, is it possible to reduce global warming enough to prevent catastrophic future harm to [People in the United States] [The stability of Earth's climate] [Future generations of people]?" Responses to each item were provided on a five-point scale ranging from "No, definitely not" (1) to "Yes, definitely" (5), with an additional option of "N/A because global warming will not cause catastrophic harm." A by-product of beliefs and risk perceptions about global warming is the overall perception of its priority for political decision-makers. To assess this, participants were asked "Do you think the following should be low, medium, high, or very high priority for the president and Congress?" with two separate items of "Global warming" and "Developing sources of clean energy." Responses were recorded on four-point scales corresponding to the question stem (low, medium, high, very high).
All of the above measures were administered both before and after participants read their assigned climate fiction story. The following section describes measures that were only administered after the stimuli because they refer to participants' experience while reading their assigned fiction story.

Narrative transportation and identification
The review above indicates that the effects of narrative persuasion can be explained by experiences of transportation and identification. To measure transportation, we consolidated the core dimensions and items of several prior lengthy transportation scales (e.g. Busselle & Bilandzic, 2009;De Graaf et al., 2012;Green & Brock, 2000;Kim & Biocca, 1997;Tal-Or & Cohen, 2010) into a short scale of three items. These three items asked "To what degree did you [have a vivid image of the events in the story] [become fully absorbed in the story] [feel as if you were present in the world that the story created]?" Responses were given on a four-point scale of "Not at all" (1), "Only a little" (2), "Moderately" (3), and "A lot" (4). An initial pilot test in an independent sample (N = 803) found that this three-item measure was internally consistent (Cronbach's α = .84) and unidimensional (exploratory factor analysis: first eigenvalue = 2.25, second eigenvalue = 0.40), and the main study corroborated these findings (α = .86; first eigenvalue = 2.35, second eigenvalue = 0.40). A full description of the methods of the pilot test and exploratory factor analysis is presented in the Supplementary Material.
Similarly, to measure identification, we synthesized the primary dimensions and items from several prior measures of character identification (e.g. Busselle & Bilandzic, 2009;Cohen, 2001;De Graaf et al., 2012;Green & Brock, 2000;Tal-Or & Cohen, 2010) into a short scale of two items. These two items asked "To what degree did you [feel the emotions the characters were feeling] [imagine what it would be like to be in the position of the characters]?" Responses were given on a fourpoint scale of "Not at all" (1), "Only a little" (2), "Moderately" (3), and "A lot" (4). This two-item measure demonstrated adequate internal consistency in the pilot test sample (Spearman-Brown coefficient = .75) and in the main study sample (.70).

Analyses
The Six Americas segmentation is calculated from the measures of worry, personal risk, risk to future generations, and personal importance (described above). A publicly-available tool (https:// climatecommunication.yale.edu/visualizations-data/sassy/) allows users to upload a dataset of responses to these four items. The tool then automatically calculates the Six Americas segment for each respondent and appends them to the dataset as a new variable. The methodological specifications of the segmentation analysis are described by Chryst et al. (2018).
H1 predicted that reading climate fiction stories would positively affect global warming beliefs and attitudes of Concerned and Cautious fiction readers. To test this hypothesis, we used an analysis of covariance (ANCOVA) for each outcome variable to compare T2 post-test means across experimental groups while controlling for the corresponding T2 pre-test variable. This "mixed" design increases statistical power relative to between-groups comparisons of post-test means that ignore pre-test values (Charness et al., 2012;Goldberg et al., 2019). In instances where the omnibus test indicated significant differences across conditions, Least Squared Difference (LSD) post-hoc comparisons were used to compare each treatment group to the control group.
H2 predicted that the effects of the treatments will be explained (mediated) by participants' reported transportation into the story and identification with the characters. To test this hypothesis, we entered the transportation index and the identification index as mediators in separate mediation models (Figure 1) using the PROCESS macro in SPSS (Model 4 in Hayes, 2013). Transportation and identification were tested in separate models because they are highly correlated (r = .69). We also used separate models for the two treatment conditions (Flight and Tamarisk), such that each model was coded as X = condition (control = 0 and treatment = 1), M = mediator (transportation or identification), and Y i = a T2 post-test outcome variable. Each model also included Y i 's corresponding T2 pre-test variable as a covariate, as in the ANCOVAs described above.
RQ1 asked whether the total (main) effects hypothesized in H1 would be observed after a onemonth delay. To inform this question, the same analyses corresponding to H1 were performed using the delayed post-test (T3) values of each outcome variable instead of the immediate posttest (T2) values. Like the tests of H1, the tests of RQ1 controlled for the T2 pre-test values of the corresponding outcome variable.

Immediate effects
Tests of H1 Results partially supported H1 (Table 1). "Flight" had significant positive effects on belief that global warming is human-caused, risk perceptions (personal risk and risk to future generations), global warming issue priority, and beliefs that climate change will cause more droughts, more poverty, more refugees, and more floods. "Tamarisk" had significant positive effects on risk perceptions (personal risk and risk to future generations), global warming issue priority, and beliefs that global warming will cause more droughts, more poverty, and more refugees. "Tamarisk" had a significant negative effect on perceptions of future flooding, likely because the dystopian future depicted by the story was characterized by a permanent drought. However, neither story significantly affected clean energy issue priority, or perceived efficacy to reduce global warming enough to prevent harm to people in the U.S. and to the stability of the Earth's climate. "Flight" had a significant negative effect on efficacy regarding preventing harm to future generations of people, potentially because the two major characters in the story agree that protesting climate change is "a waste of time" (Simpson, 2011, p. 9) given its advanced stage.
The effects of the two climate fiction stories were very similar to each other. The LSD post-hoc pairwise comparisons indicated that there are significant differences between participants in the Flight and Tamarisk conditions on only three outcome variables. The first of these variables was belief that global warming is human-caused, which was significantly higher in the Flight condition than the Tamarisk condition (M diff = 0.15; SE diff = .035; p < .001). This may be because the characters in "In-Flight Entertainment" explicitly discussed the reality of global warming and one character presented rebuttals to common counter-arguments. A second instance of significant differences between treatment groups was in the belief that global warming will increase floods, which was lower in the Tamarisk condition than the Flight condition (M diff = 0.36; SE diff = .036; p < .001). This is likely because "The Tamarisk Hunter" was set in a parched dystopian landscape where water was a scarce commodity, making floods seem less likely. A third instance was the efficacy perception regarding preventing harm to future generations, which was lower in the Flight condition than the Tamarisk condition (M diff = 0.15; SE diff = .046; p < .001).
As an exploratory analysis, we investigated whether gender, age, political ideology, and Six Americas segment moderate the treatment effects. The tests of these interactions revealed that the treatment effects did not differ significantly across levels of any of these potential moderators. These results are presented in the Supplementary Material.

Tests of H2
The immediate post-test also supported H2, such that for both the Tamarisk and the Flight treatments, participants' level of transportation and identification (respectively) each mediated several of the significant X-Y relationships. However, most of these significant indirect effects involved partial mediation, such that the residual direct effect of X on Y was still significant when accounting for the mediator. The only instances of full mediation of a significant main effect were, respectively, transportation and identification's mediation of the effect of "The Tamarisk Hunter" on perceived personal harm. In these results, transportation seemed to consistently be a stronger mediator of the effects of "In-Flight Entertainment" than identification (Table 2). This may be explained by the fact that the "In-Flight Entertainment" (relative to the control) had a larger effect on transportation (β = .54, SE = .07, 95% CI [.41, .67]) than on identification (β = .17, SE = .07, 95% CI [.04, .31), Z = 3.89, p < .001 (for the formula used to compare regression coefficients, see Paternoster et al., 1998). Regarding "The Tamarisk Hunter," the indirect effects of transportation and identification were similar across most outcome variables. This is, in part, because the "The Tamarisk Hunter" (relative to the control) did not have a significantly different effect on transportation (β = .48, SE = .07, 95% CI [.34, .62]) compared to identification (β = .51, SE = .07, 95% CI [.37, .65]), Z = 0.34, p = .733. In theory, it is possible to evaluate the unique indirect effects of each mediator while accounting for the other (e.g. via parallel mediation). However, as mentioned above, transportation and identification were not included as mediators in the same models due to high intercorrelation.

Tests of RQ1
The analyses revealed that after the one-month delay, there were only two significant effects of reading a climate fiction story, neither of which were positive (Table 3). At Time 3, participants who had read the Tamarisk story reported significantly lower belief that global warming is happening and expected significantly fewer floods due to global warming, compared to those in the control condition. All of the significant positive treatment effects that were observed in the immediate posttest had diminished to statistical nonsignificance. However, this is not conclusive evidence that these effects disappeared entirely, because the difference between the significant effects and nonsignificant effects was not significant (see Gelman & Stern, 2006) (Figure 2).

Discussion
Because climate change fiction is a growing genre of literature and much research indicates the persuasiveness of narratives, this study conducted the first randomized controlled experiment on the persuasive effects of reading a climate fiction story. The results indicate significant small effects on a set of beliefs and attitudes that in most cases were partially mediated by an experience of narrative transportation and in some cases were partially mediated by identification with the story's characters. These effects were observed immediately after the treatments, but decreased to nonsignificance after a one-month delay. While the two climate fiction stories were very different in content and style, their respective effects were very similar in size and transience, which may suggest some tentative implications about generalizability to climate fiction short stories at large. These findings build on a large body of literature indicating the persuasive power of storytelling in environmental issues, and they add important detail about the longevity of the effects that occur from a single exposure to a message. Further research is needed to test other important questions such as how different genres, plot elements, and styles of climate narratives might have different effects. Most empirical research on the influence of literary narratives has used short stories or excerpts, but a novel might have larger and longer-lasting effects since its greater length and immersive detail could cause greater transportation and identification and provide more information about environmental issues. However, one difficulty of generating knowledge about the mechanisms of narrative persuasion is that narratives are typically long and complex, and their many elements and idiosyncrasies make it difficult to ascertain which aspects are driving the observed effects.
In addition, readers' literary preferences, prior worldviews, demographics, and psychological traits may each affect how they process and respond to literary works with environmental themes. While the present study focuses on a specific, practically important subset of the population (American fiction readers in the Concerned or Cautious segments of Global Warming's Six Americas), future research could investigate the effects of reading climate fiction among other subpopulations.
One limitation of this study is that many readers of climate fiction likely belong to the Alarmed segment, not the Concerned or Cautious. We excluded the Alarmed from the present study because they already score very high on the outcome variables, indicating that they do not need to be persuaded that climate change is a significant threat. Still, future research should investigate the effects of climate fiction among the Alarmed. Prior research has indicated that narratives can strengthen the attitude-behavior relationship (Rhodes et al., 2016), so reading climate fiction may help activate pro-environmental behavior in individuals who already have strong pro-environmental attitudes. Another important consideration when interpreting these results is that Mechanical Turk samples are not representative of the broader population. We recruited from this pool of participants because it enabled us to study a very specific sub-population, to cover a wide geographic range of the U.S., to recontact participants after a long interval, and to recruit participants into T2 and T3 without revealing that it was associated with a prior survey they had completed (thus reducing the risk of selective attrition). Despite these advantages, the non-representative nature of this sample is a barrier to generalizability. It is also important for readers to consider the implications of our findings of transient effects. The immediate treatment effects observed in the present study were small-to-medium in size at the immediate post-test measurement and were nonsignificant after a one-month interval. As much prior research has shown, immediate changes in response to a single stimulus usually do not result in permanent changes. However, this distinction should not be interpreted as evidence that these messages failed, or that climate fiction can only have small and transient effects on readers' beliefs and attitudes. For instance, research indicates that repeated exposure and reinforcement from other media and social influences can lead to compounding effects and to behavioral change (e.g. Abelson, 1985;Funder & Ozer, 2019). For example, Carnahan et al. (2020) demonstrate that the effects of a single message tend to dissipate, but repeated messages can cause the effects to stick. One of the limitations of the study is its single-exposure message design, but finding significant immediate effects from a single exposure in an artificial setting is still a valuable insight. In part, this is because the effects of a single exposure in an artificial setting may represent a lower bound of the real-world effects. Reading climate fiction in the real world often involves multiple exposures and longer narratives, which may result in larger and longer-lasting impacts.
The findings of this study have widespread practical implications. First, it highlights the effects that the growing genre of environmental literature is likely having on the beliefs and attitudes of its readers. Further, educators, researchers, publishers, and activists seeking persuasive climate change communication content should note the growing body of research indicating the effectiveness of narratives and storytelling, and of climate fiction specifically. There are many existing works of climate fiction and non-fiction storytelling that could be used in diverse media and contexts. This study also finds that the effects of these stories are often explained (at least in part) by felt transportation into the story and sometimes by felt identification with the characters. Thus, a practical recommendation for strategic communicators is to seek to maximize these two experiences when making decisions about message format, style, and content.
This study opens up several new research directions as mentioned above, and we hope that this empirical test of the effects of environmental literature will motivate further interdisciplinary collaborations between critical scholars and empirical researchers.

Disclosure statement
No potential conflict of interest was reported by the author(s).