In this study, 64 participants with a self-reported active ED read either an unpublished ED recovery memoir or a control text unrelated to EDs over a roughly two-week period and completed the EDE-Q and ANSOCQ one week before and two weeks after reading, as well as responding to three repeating open-ended questions at regular timepoints during and after reading. The intention was to establish whether any change in ED severity was manifested after versus before reading in order to guide a yes/no decision as to whether to publish the memoir. We also sought to investigate the phenomenology of the reading process as it relates to the processes of illness and recovery and individuals’ attitudes to them, to enhance understanding of these response dynamics as well as to guide pre-publication edits to the recovery memoir.

Clinical measures

In this pre-publication study of readers’ responses to an ED-themed memoir and ED-unrelated control text, both ED symptomatology (as measured by the EDE-Q) and ED attitudes (as measured by the AN-specific Stages of Change Questionnaire) were improved for post-reading versus pre-reading for both the experimental (HA) group and the control (TZ) group. Both groups manifested significant positive change with effect sizes that were moderate-to-large (for the EDE-Q) or very large (for the ANSOCQ).

These findings run counter to widespread intuitions (for example amongst many clinicians) that in general reading may have minimal effects on ED-relevant factors, perhaps especially core symptomatology. They also contradict existing evidence suggesting that engagement with texts may have modest or no detectable positive effects on ED-relevant dimensions, including Riestra-Camacho, Carney, & Troscianko’s [9] study of young-adult sports fiction and much research in self-help bibliotherapy. In self-help bibliotherapy studies, even self-help books designed for the sole purpose of eliciting therapeutically valuable responses typically generate only minor to moderate positive effects, with research remaining inconclusive about many basic questions of efficacy and mechanisms. The results also diverge from the strong retrospective self-report evidence in Troscianko’s [5] survey study that ED-specific narratives can have rapid and significant harmful effects on core ED dimensions. Finally, the results also challenge the “similarity thesis” (the expectation that therapeutic ED-related effects depend on thematic congruity between the textual content and the reader’s health situation) since here an effect was found only of time and not of condition—that is, engagement with the ED-themed and the ED-unrelated text had equally positive effects.

The study’s main findings are compatible with a popular belief that engagement with narrative or “literary” texts exerts strong effects on readers—effects typically assumed to be positive in nature. This belief underpins narrative bibliotherapy, which seems to be a relatively common informal practice in one-to-one therapeutic settings, and it is implicit in much humanities research and many public-facing literary arts initiatives. However, the results do not unproblematically support the thesis that “books do good”, because it is unclear which features of these two texts may be primarily responsible for the observed improvements, or indeed to what extent the textual features themselves were a major driver at all.

Beyond the impact of text, the absence of mediating effects found in the ANCOVA for either the EDE-Q or the ANSOCQ of any of the main continuous variables (age, professional support duration, highest educational level, BMI change) suggests that the main effects are not significantly susceptible to contributions from demographic or ED severity/support variables, nor indeed from the text choice itself. This null finding further supports the idea that the benefits derive from facets of the text/reader interaction that have yet to be confirmed.

In the remainder of this section, therefore, the free-text response data will be used to generate provisional conclusions as to the mediators of the two main effects and related phenomena. We begin with the computational analysis on three analytical dimensions, offering suggestions as to the drivers of effects related to emotional variance, somatosensory effects, and text/response similarity. We conclude by offering insights generated by close reading of participants’ reactions to dialogue form and the experimental setup. Unedited illustrative quotations from participants’ responses are included as subsection epigraphs and as part of the main discussion. Unless otherwise indicated by bracketed ellipses ([…]), all quotes represent the entirety of a participant’s response to a given question.

Computational analysis of free-text responsesVAD dimensions

“I started this push at recovery a year ago, and feel it has failed. I can’t believe how hopeful I felt back then. I feel so ashamed at how I’ve behaved in the interim; it seem perverse that recovery should have destroyed my relationships more effectively than anorexia. I don’t think I nor my family had expected quite how much anger and fear and emotionally-charged reforging of me would need to be done - we’ve all been shocked. Pity is easier to give, if not to receive. What do you give someone who is explosively recovering?” (HA participant)

Using the VAD norms to analyse participants’ free responses (Fig. 3) found a significant within-participants effect for arousal, which increased across the six reading timepoints, and no effects for valence or dominance. The increase in arousal was especially marked for timepoint 6 in HA responses, where an unexpected death and mourning process are described. This thematic shift intensifies the general expansion out from ED-centric themes and experiences that characterizes the later stages of ED recovery, as noted by one HA participant: “Different than the previous sections. More about other life challenges than food-specific ones, which would reflect life after recovery.” Three HA participants at timepoint 6 used language that is particularly rich in high-arousal words, including “rollercoaster”, “climax”, and “inspired and scared”. Arousal increased more steadily across TZ responses between timepoints 3 and 6, and there were no striking examples of high-arousal language in the TZ free-text responses.

Arousal and dominance are typically inversely correlated, but in responses to both texts, the dominance as well as the arousal peak was observed at timepoint 6, suggesting that the narrative progression in both texts—and/or the processes involved in responding to them—had both arousing and control-enhancing elements. The absence of significant change over time in valence values in both groups, and free-response mentions of enjoyment or otherwise that demonstrated no obvious pattern either, suggest that liking and dislike did not play a major role in any reading-related effects. This aligns with findings from a group-reading setting, in which liking or disliking the text being read did not correlate with the perceived value of participation, and liking or disliking the group discussion did not usefully capture the emotional variation in these sessions [33]. One participant whom we contacted because she seemed to be finding the reading emotionally difficult clarified that although she was not enjoying the book very much and it would not be her preference stylistically, the reading and question-answering was encouraging meta-reflection on her thought processes and greater honesty about her current situation. The VAD results and comments of this kind further encourage an interpretation of the main findings according to which the processes of engagement trump the textual features in driving effects in readers—and adds the further hypothesis that enjoyment of these processes is not a major predictor of their value.

Sensory and action-effector dimensions

“Confident and exhausted at the same time. Exhausted because everyday I ‘get something wrong’: I don’t manage to stop bingeing, I can’t force myself to eat afterwards, or because I’m afraid to challenge myself. Confident becase sometimes, when I give up controlling and micromanaging, wonderful things happen, giving me hope. Yesterday I had a proper and elaborated afternoon snack (not only an apple, but yogurt with rice cakes and peanut butter, and fruit) and unexpectedly I stopped eating when I felt ok. No binge eating on peanut butter as usual.” (HA participant)

As for the sensory and action-effector dimensions, the increase in gustatory values between timepoints 2 and 5 in the experimental group is explicable with reference to the increasing textual emphasis on food and eating. HA naturally contains a much higher density of food-related terminology than TZ. A number of HA participants mentioned being bored, overwhelmed, or otherwise put off by the amount of food-related language in the middle and later sections of the book; these reactions may partially account for the increase in gustatory language for this group.

Olfactory and mouth norms tend to track gustatory ones; concreteness and imageability, which tend to co-vary, may also be driven by the gustatory language. The marked increase on visual ratings at timepoint 6 in both groups suggests a possible effect of the use of visual metaphors for increased cognitive “insight” or a more “zoomed-out” perspective at the end of the two books, corresponding to textual elements in both that convey a trajectory of culmination. The similar pattern for “head” may also be tied to the use of visual metaphors for cognition. The decreases in foot/leg and torso at timepoint 6 are not readily accounted for, but may be inverse correlations with an increase in cognitive insight. Body-related language mostly concerned difficulties with body image and bodily sensations, worries about fat, and other negative body-related feelings and attitudes common in ED experience.

Text/response similarity

“Interesting. It’s a challenging read and different to what I’d usually go for. But I’m getting a lot from it, and find the writer’s voice engaging and very clear. As she’s struggling to come to grips with some of these very confusing questions, she’s bringing us, the readers, with her on her journey. Am I conscious now? And now? It really is like waking up as soon as the question is asked. Very interesting!” (TZ participant)

Analysis of similarity between the texts and responses to them found that the responses for TZ most closely matched the text of TZ, and the same was true of HA, indicating that both texts significantly shaped the language choices of the participants answering questions subsequent to reading them. Without a linguistic corpus to provide a baseline for divergence, relative strengths of these effect for HA versus TZ are hard to assess, but a possibly stronger effect for HA may reflect a greater thematic capture of ED-related material, as suggested by the higher frequency in the TZ group of comments to the effect that they had found the text boring to read or struggled to concentrate on it. Specifically, 15 HA vs. 23 TZ participants mentioned being bored or not engaged or put off at one or more points in the text. These responses may reflect the difficulty of concentrating on non-ED-related stimuli when cognition is impaired by an ED; and this kind of “cognitive constriction” [5, p12] may be especially marked when an ED is associated with malnutrition [34]. As one might expect, valuing cognitive/emotional exploration showed the inverse pattern from boredom/disengagement across the two groups, with more comments to this effect in HA than TZ (20 HA versus 14 TZ). Overall, close reading suggests that TZ was somewhat more polarizing than HA in terms of liking or enjoyment, perhaps in part because of the lower cognitive-emotional capture of non-ED material. These differences, combined with the overall trend towards lower values for TZ than HA on the sensory and action-effector dimensions, suggest that ED-themed material may have a more directive channelling effect on cognition, offering some support for the cautionary angle predicated on the intuition that these texts in some sense “land harder”.

Close reading of free-text responses: textual and contextual factors

Beyond the three types of computational analysis, close readings of the free-response data carried out by ET identified responses to the dialogical form of the two texts and responses to the experimental setup as possible mechanisms mediating the substantial positive effects of time but not text.

Responses to dialogue and question/answer structures in the texts

“I feel very grateful to have taken part and I think that the concept of the book is great although for me personally it did not work well. It is amazing and intense to get the insights straight in direct speech. […]” (HA participant)

We suggested in the Results section that participants’ responses to the dialogical and question-and-answer form may have contributed to the positive main effects observed. These responses in the HA group reflected, to varying degrees, all four of the potential effects of this rhetorical choice listed in the Introduction:

The alternating expression of and challenges to the ED perspective (e.g. “[…] In comparison to other memoirs I have read the ‘B’ voice acted as a kind of ‘leveller’ or reminder of rationality in the conversation. […]”)

The enhancement of insight through dialogical interplay (especially as a three-way interaction involving the reader, e.g. “[…] as I read the questions [posed by “B” to “A”] I tried to answer them myself. As I am now in the middle of recovery, whereas when I first started reading section one I was deep in my anorexia, I could feel the way my anorexic self would have answered the questions, and compare it to how I would approach them now. […]”)

The distinct roles of dialogical reflection versus behavioural change (e.g. “[…] i hadn’t realised until right at the end that A and B (and C) were different voices in her head. I feel like that was a good way to explore how stuck in your head you can be in the depths of your eating disorder. […] // […] Emelia was now whole again, it was one monologue rather than multiple voices, and her life was so much bigger than food. But her life wasn’t perfect - she still had problems, she still had issues with her body image some times and comparisons. But it was great that she didn’t fall back into bad habits.”)

And the possibility of recovery despite ambivalence (e.g. “[…] The increased coming together of A and B towards the end of the section (them sharing more certainties) was very reassuring, though, in that it encapsulates the shifts of identity that seem so scary right before recovery and that are actually a lot easier to accept (and welcome) and feel more natural once recovery is taking place. […]”)

One way in which the dialogical and interrogative form may have contributed to beneficial effects for participants is in allowing for a less passive readerly interaction than a more traditionally constructed memoir. One HA participant noted at timepoint 1: “[…] The question-answer format of this section sort of ensures that it be necessarily more thought-provoking than most anorexia recovery books, and demands more active (and self-critical) engagement from the reader, which I found to be very interesting and useful in terms of imagining my recovery.” This suggests a role for empowering alternatives to the typically didactic nature of the self-help book, which can be paradoxically dis-enabling (D. Holloway, personal communication). At timepoint 2, the same participant continued: “[…] I suppose reading this section while in recovery sort of recreates the form of the text itself (the splitting of the speaker into A/B/C etc). Overall I’ve found it quite a therapeutic experience, in that it’s been a bit like performing therapy on myself.” Interestingly, this sense of enhanced agency in using the text for personal growth was experienced also by TZ participants. One made the comparison with self-help books explicit, emphasising specifically the greater autonomy invited here: “[…] I’ve appreciated reading a book that is not the typical self-help book, where you’re given a templet of things you should do, think and feel: first do A, then B, then C. If I haven’t completely misunderstood this book, it opens up to bigger, more philosophical questions (rather than more practical ones), which the reader can think about and figure out herself. And I appreciated that.”

Overall, the reports of significant cognitive demands made by both texts indicate that readers found the interpretive process challenging—in some cases off-puttingly so, in others enjoyably, but perhaps in either case as a major contributor to important forms of cognitive-emotional exploration and meta-reflection. Participants’ responses testified to the complexity of their responses, in which “positive” and “negative” cannot be neatly separated out, and in which difficult or uncomfortable experiences often form part of a broadly enriching process. This characteristic aligns with the group-reading findings alluded to earlier [33] in which liking and enjoyment did not contribute substantially to the perceived value of participation, and offers a way of accounting for readerly benefits (as indicated by the large effects on the two standardized measures) that does not depend on an anodyne concept of bibliotherapeutic change. The editing process prior to publication will preferentially target the elements of confusion in HA that can be reduced without compromising the intended ambiguities and complexities.

Responses to experimental setup

[Please, wait until tomorrow to carry on reading. Thank you.]

“Roger that. Thank you for these little snippets of kindness and curiosity. Cheers.” (TZ participant)

As we ask what interactions of textual features and cognitive processing may be driving the EDE-Q and ANSOCQ improvements, it is also possible that the text itself is something of a red herring. The nature of the reading and interpretive processes themselves may exert stronger effects than the linguistic content on which they centre. This would converge with the findings in Carney & Robertson [12] showing that positive effects of reading fiction on mood and wellbeing may emerge only once readers are given the opportunity to reflect on the text they have read, whether through recall or discussing texts with other readers. That is, merely reading a text is not enough for the benefit of the text to be actualized. In the present study, participants were invited to reflect on HA and TZ via staged questions at regular timepoints, thereby providing a context for reflection that was at once sustained and semi-social. On this view, the otherwise puzzling presence of an effect for time and no effect for textual condition becomes explicable as the result of giving readers an ideal scenario for reflective consolidation, where the fact of consolidation is more important than the details of the texts being read. This account also aligns with the “absolute sleeper effect”, in which readers continue to think and be influenced by what they have read in a book long after finishing it [35].

Such results raise the further possibility that the methods designed here to elicit free-response testimony on the reading experience and its effects and context may have substantively changed the processes under investigation. This is to some extent inevitable with any investigative procedure: Volunteering for a research study and reading a book in a specific format, within a specific timeframe, accompanied by a range of self-report tasks, cannot leave the reading experience unaltered relative to recreational reading. Reactivity effects are well documented in psychology and health contexts, with the fact of being observed or measured generally increasing positive behaviours and reducing negative ones (for an overview, see [36]), and self-tracking showing similar positive effects [37]. The effects of the qualitative aspect of the study design may be wider-ranging than expected, however. The value of responding to recurring questions throughout the reading process was mentioned by several participants, including as transferable to everyday life, while others disliked certain aspects of the question-answering procedure or suggested alternatives, and others again used the questions as prompts to what we might identify as self-coaching.

Our first recurring question, “How is your day going?”, functions similarly to the most basic of journaling prompts used by millions of individuals worldwide every day. Some researchers on expressive writing and its therapeutic benefits have suggested that exploration of self through writing is less meaningful if it does not encompass a wide temporal span taking in the moderate to distant past. In a diary-writing study, for example, Green [38, p142] reports that the group who received a long Wordsworth poem to reflect on in their writing showed evidence of a broad range of memories being elicited by the text, whereas the control group’s entries “concentrated overwhelmingly on the daily routines and habitual preoccupations of each participant”, with the strong implication that the control group’s writing was therefore less meaningful or valuable. But the search for the origins of illness may often be a misguided endeavour in the context of mental health [39] and a cognitive-behavioural perspective would suggest that the benefits to be derived from conceptual/linguistic interrogation of one’s self and life are likely to accrue predominantly by shifting the interactions between present and future rather than by taking up a new explanatory stance on the past. The present study’s findings are compatible with the idea that meaningful insights and the potential for meaningful change can arise from prompts and focal points that are not primarily past-oriented, and via textually inflected reflection that does not involve significant amounts of recall or interpretation of distant memories. Overall, the combination of open-ended prompts to reflection plus the communicative structure of respondent and recipient may have contributed to a process with qualitatively different interpretive outputs relative to reading a text with only the pre- and post-reading questionnaires.

Limitations and future directions

We have speculated as to the possible mechanisms underlying the main effects found in this study: the significant improvement for both groups on the EDE-Q and (with an especially large effect size) the ANSOCQ. We have identified potential contributors amongst contextual factors in the experimental setup (specifically the recurring open questions) and features of textual construction (dialogue form). Insights yielded by other aspects of the free-response data—on dimensions including triggering, cognitive-emotional exploration, and identification and related phenomena—will be reported in a separate publication. More research will be needed to establish causal dynamics with confidence, particularly with respect to the question of the relative contributions of textual versus contextual factors. Resources permitting, inclusion of an assessment-only condition, in which two of the three recurring questions (excluding the reading-focused one) would be administered with no reading task, could shed further light on this question in future studies. On the question of thematic emphasis, further inquiry is called for into the distinction between ED-themed and non-ED themed narratives, which previous observational data [5] suggested is fundamental but whose significance the current experimental findings call into question.

In the present study, we used two unrelated texts as experimental and control text, cognisant of the fact that there is no such thing as either a perfectly matched text (any given pair of texts will vary on multiple dimensions) or a perfectly neutral text (every text has complex and partially unpredictable effects on multiple dimensions for different readers at different times). The texts were significantly different in length (HA was much longer), which may have created uncontrolled-for variance in responses. In previous work [26], we used one text in two versions, differentiated by the changes that the author (Franz Kafka) himself made during the writing process, and argued that this method offers a strong combination of comparability and ecological validity. This method is often not viable for texts that are not equipped with the extensive scholarly apparatus characteristic of the Western canon. In practice, other experimental/control combinations are also needed, and future investigations into health-related effects of reading will depend on generating appropriately robust but realistic paradigms.

Using a control group for reading studies can pose difficulties specifically in relation to participant expectations and demand characteristics. Here, HA participants received the type of book they were expecting, while TZ participants were not expecting to read a book about meditation and so may have reacted to it in ways that would be uncharacteristic of individuals who had chosen to read such a book—for example, reading faster, skipping bits, or reading with less interest altogether. Some TZ participants indicated that this had been the case in their reading. Specifying in the recruitment materials that meditation would be the control text subject matter might have helped reduce problems with participants finding TZ boring or dropping out altogether. (The TZ group had 63 dropouts versus 53 in the HA group, though mostly before beginning to read.) However, some level of disappointment not to be assigned to the experimental group may be inevitable.

In future studies we would advise including an initial reading checkpoint at the very start of the text, in order to generate more accurate data about total reading duration. We estimate that the mean duration here was approximately one month, but this involves rough estimates of the gap between the start of reading (as opposed to when text access was provided) and the first checkpoint being reached. There were 3 outliers who took a long time to complete the reading: one TZ participant who took around 10 months, and two HA participants who took three and four months respectively. Data collection for this study took a total of 13 months, a relatively long duration that may have reduced data consistency both at an individual level (with more potential for changes in life circumstances and events) and across the groups (especially considering that the year between summer 2021 and 2022 was a year of significant change globally in relation to the progression of the Covid-19 pandemic).

The post-reading measures were taken two weeks after reading, aiming to avoid measuring too soon and failing to capture the persistent and potentially increasing aftereffects of reading complex texts [35] and also measuring too late to capture possible short-term “triggered” effects. The timing of the post-measure meant that the full two suggested weeks of reading would be captured in the questionnaire responses. We opted not to include a longer follow-up phase in order to avoid amplifying recruitment challenges, but in future studies a light-touch follow-up after approximately three months would be advisable to test for maintenance or loss of the post-reading effects. Future studies should take into account the possibility that initial positive effects may be short-lived or conversely that short-term negative effects may be precursors to longer-term benefits, potentially via uncomfortable confrontation of difficult realities.