1. Linguistic Analyses of Natural Written Language: Unobtrusive Assessment of Cognitive Style in Eating Disorders Markus Wolf, Jan Sedway, Cynthia M. Bulik, Hans Kordy ABSTRACT Objective: Linguistic style analysis is introduced as a method of nonreactive measurement of cognitive style in individuals with eating disorders (ED). Method: A journaling exercise was implemented on an ED inpatient unit. Thirty-four written essays collected from an unselected sample of 11 patients were analyzed with the Linguistic Inquiry and Word Count, with regard to predefined text categories. At each session, self report of participant well-being and session impact were assessed. Results: Compared to essays from a student control group, and message-board entries of individuals who had recovered from an anorexia nervosa (AN), the inpatient journals displayed the highest rates of self-related words, negative emotion words, and the lowest rate of positive emotion words. Inpatients used more anxiety words and fewer words that refer to social processes and eating concerns than individuals who had recovered from an AN. Associations were found between linguistic categories, prewriting well-being, and post writing evaluation. Conclusion: Linguistic style analysis offers insights into cognitive styles, and provides a promising approach for their unobtrusive measurement in ED. VVC 2007 INTRODUCTION Therapeutic writing and journaling exercises are helpful interventions in psychotherapy and psychosocial care. Moreover, therapeutic writing has been suggested as an adjunct treatment in the inpatient treatment of anorexia nervosa,4 as a means of circumventing avoidance behaviors and difficulties in disclosing personal thoughts and feelings. Recently, researchers have moved beyond writing as an intervention by applying of text analytic methods to quantify writers’ inner thoughts, feelings, and cognitive strategies. For example, Rude et al.7 applied a computer-assisted text analysis program, the Linguistic Inquiry and Word Count to analyze written essays from students with varying levels of depressive symptoms. They identified what had previously been labeled the ‘‘language of depression.9’’ Lyons et al.10 explored linguistic markers in Internet self-presentations of individuals who adhere to a ‘‘pro-anorexia’’ lifestyle. Analyzing message-board entries with the LIWC, they reported distinct language patterns of pro- anorexics in comparison to individuals recovering from anorexia nervosa with regard to emotional and cognitive processes, general social versus self-awareness, anorexia-related concerns, and temporal focus. Pro-anorexics displayed more positive emotions, less anxiety, a lower degree of cognitive reflection, and lower levels of self-directed attention than those in recovery. Furthermore, they were more focused on the present and were more preoccupied with eating. Amanda Jordan METHOD Participants The project was conducted on a specialized unit where intensive inpatient treatment is provided for patients with an ED. A journaling exercise was implemented in late 2005 as part of the comprehensive treatment program. The sample was predominantly women (n 5 10) and Caucasian (n 5 9); two participants were Asian. Their mean age was 35.5 years (SD 5 15.4; range 18–56 years). Diagnoses were determined at admission by semi structured intake interviews designed to assess DSM-IV EDs. All intake assessments were conducted by psychiatrists or clinical psychologists with ED expertise. Instruments Prior to each journaling session, patients were asked to rate their current well-being with the following question: ‘‘On the following scale, please indicate how your general well-being is today.’’ This scale is similar to a visual analogue scale, with higher values indicating better wellbeing (item range from 0 5 ‘‘very poor’’ to 10 5 ‘‘very well’’). We developed the well-being scale for the quick assessment of patients’ level of prewriting distress over the course of the inpatient treatment. The text analyses were conducted with the LIWC.8 The LIWC is a dictionary-based text analysis program, which characterizes text with reference to 74 standard linguistic (e.g., word count, words per sentence, first person singular, past tense), psychological (e.g., emotion words, cognitive words, social words), and content (e.g., eating, sports, death) categories by counting the words that refer to each of these categories. For each category, results are displayed as mean percentages of words found in any given text, except for word count and words per sentence. On average, the LIWC dictionary captures 80% of the words used in writing.8 The program has been extensively validated and applied.6 Data Preparation and Analysis The hand-written journals were transcribed, spellchecked, and stored electronically in the appropriate text format for further LIWC analyses. Each journal was treated as a unit and processed separately. RESULTS A total of 34 essays were written by the 11 participating patients. Four patients participated in the journaling exercises three times or more, three patients participated twice, and four patients participated once (Fig. 1). On an average, an essay consisted of 603.6 words (SD 5 259.1), with a range of 96–1,125 words. High hit rates were found in the text analyses, with the LIWC dictionary capturing on an average 81.5% of the written words (range 73.3%–92.7%). CONCLUSIONS We analyzed the language of individuals with an ED through computer- assisted quantitative text analysis. Our comparison of journaling essays written by patients with ED over the course of their inpatient treatment with similar scripts from individuals recovering from anorexia nervosa and student controls yielded results consistent with our hypotheses. In the essays of ED inpatients, we found the highest rates of self-related words, negative emotion words, and past tense words, as well as the lowest rates of positive emotion, present tense words, and words associated with eating or death and dying. The differences were highest between the ED inpatients and the student controls, while individuals recovering from AN scored in between those two groups. LIMITATIONS AND RECOMMENDATIONS The results of our study require replication with a larger sample. Our study suffered from low levels of participation and high drop out. Similar experiences were reported by Schmidt et al.,4 who noted difficulties in establishing a writing exercise for AN inpatients on their unit. One possibility is that cognitive function is significantly impaired in individuals who are at extremely low body weight, rendering journaling a difficult task.