Adriel Boals, Kitty Klein, & Allyson Banas

There is now a great deal of evidence (Smythe, 1998) that people asked to write about stressful experiences enjoy improvements in health (Pennebaker & Francis, 1996), and immune system function (Pennebaker, Kiecolt-Glaser & Glaser, 1988; Petrie et al., 1995) compared to individuals assigned to write about trivial topics. Current theories (Pennebaker, 1997) emphasize the role of narrative coherence as the means whereby expressive writing produces health effects (Esterling et al., 1999). Pennebaker, Mayne & Francis (1997) operationally defined narrative coherence as the relative frequency of cognitive words related to cause and insight. The present experiments were designed to test the validity of Pennebaker et al.'s coding scheme and then compare the coding scheme to an alternative system for coding narrative coherence. In the first experiment, we compared thirteen popular songs that are strongly narrative with thirteen songs that were not developed narratives. In the second experiment, we measured narrative using a new holistic coding system (Katz, 1999) and compared the results to those given by Pennebaker et al's coding system.

Lyrics from 13 popular songs that were judged to tell a story about a past event (Narrative Songs) and 13 popular songs that were judged to not tell a story (Non-narrative Songs) were chosen (see Appendix) and scored for narrative coherence using Pennebaker et al's Linguistic Inquiry and Word Count (LIWC) program. LIWC is a software program that contains word categories, and counts the percentage of words used in an essay for each word category. Narrative coherence is defined as the percentage of words used that are considered "cognitive mechanism" words such as "cause", "know", and "ought".

Contrary to expectations, t-test analysis found that non-narrative songs contained a larger percentage of cognitive mechanism words (M = 7.7) than narrative songs (M = 4.5), p = .01. Results also revealed that non-narrative songs contained a larger percentage of insight words (a subcategory of cognitive mechanism) (M = 3.1) than narrative songs (M = 1.5), p = .02.

The results from Experiment 1 suggest that a count of the number of cognitive mechanism words used in a writing sample may not be an accurate measure of narrative coherence. In Experiment 2, we designed a holistic system for coding narrative coherence and compared it to results of a LIWC analysis.

Sixty-eight students were randomly assigned to write three times for 20 minutes per session about either a very positive (n = 32) or very negative personal experience (n =36) using standard expressive writing instructions (Pennebaker & Francis, 1996). The writing sessions occurred within a 2-week period during the spring semester. We obtained students' permission to access their health records from the Student Health Services.

The first and third essays were coded using Katz's (1999) system. Narrative coherence was defined as a function of the extent to which an essay includes a character, a scene, a beginning, a middle, an end, and has consistent topics. The material scored were essays written by students in an expressive writing experiment. The ICD-9 billing codes were applied to the students' health center visits. To measure health outcomes, we used the number of visits for which the diagnosis was an infectious disease. Visits were categorized as made the 6 weeks before the writing experiment began, the 6 weeks following its completion, and the subsequent fall semester.

Two coders were trained on the Katz system. For the first essay, agreement between coders was .71; for the third essay, agreement was .80. As was the case in Experiment 1, higher narrative coherence was associated with lower use of cognitive mechanism words in the first, r (66) = -.38, p < .002) and the third essays, r (66) = -.27, p < .03. The health data obtained before the experiment began was used to calculate residual scores for the period immediately following the experiment and the subsequent semester. These residual scores were used as dependent measures in a Writing Condition X Coherence Score repeated measures ANOVA. Significant interactions between changes in narrative coherence and condition, F(1,64) = 5.52, p < .02, indicated that for students who wrote about negative experiences, higher coherence scores were associated with more health center visits than predicted. For students who wrote about positive experiences, higher coherence scores were associated with fewer than predicted health center visits.

The data indicate that contrary to expectation, two traditional measures of narrative coherence are inversely related to coherence defined as words indicative of cognitive mechanisms. In Experiment 2, increases in narrative coherence predicted poorer health outcomes for as long as 8 months following expressive writing about negative experiences. Defining narrative coherence using a traditional literary approach produces a very different pattern of results than defining narrative as a simple increase in the use of cause and insight words. The discrepancy has implications for current theories about how expressive writing produces health benefits and how memories of stressful experiences are transformed through retelling.

Esterling, B. A., L'Abate, L., Murray, E.J., & Pennebaker, J. W. (1999). Empirical foundations for writing in prevention and psychotherapy: Mental and physical health outcomes. Clinical Psychology Review, 19, 76-96.

Katz, S. (1999). Training manual for coherence scoring. North Carolina State University Department of English, Raleigh, NC.

Pennebaker, J. W. (1997). Writing about emotional experiences as a therapeutic process. Psychological Science, 8, 162-166.

Pennebaker, J. W., & Francis, M. E. (1996). Cognitive, emotional and language processes in disclosure. Cognition and Emotion, 10, 621-626.

Pennebaker, J. W. & Francis, M. (1999). Linguistic Inquiry and Word Count: LIWC. Mahwah, NJ: Erlbaum

Pennebaker, J. W., Kiecolt-Glaser, J. & Glaser, R. (1988). Disclosure of traumas and immune function: Health implications for psychotherapy. Journal of Consulting and Clinical Psychology, 56, 239-245.

Pennebaker, J. W., Mayne, T. J. & Francis, M. E. (1997). Linguistic predictors of adaptive bereavement. Journal of Personality and Social Psychology, 72, 863-871.

Petrie, K. J., Booth, R. J., Pennebaker, J. W., Davison, K. P., & Thomas, M. G. (1995). Disclosure of trauma and immune response to a hepatitis B vaccination program. Journal of Consulting and Clinical Psychology, 63, 787-792.

Smythe, J. M. (1998). Written emotional expression: Effect sizes, outcome types and moderating variables. Journal of Consulting and Clinical Psychology, 66,174-184.

For questions or comments, email

Adriel Boals