LIWC and the TAT

Abstract

I led the data preparation and analysis for a collaborative Thematic Apperception Test project at Fielding Graduate University, scoring 677 card responses from 69 participants across 31 cards with Linguistic Inquiry and Word Count. Across the five most frequent cards, respondents used more of the language each card was designed to elicit; stories about Card 8BM contained 3.18 times the corpus-average rate of death language, and stories about Card 3BM 2.74 times the rate of sadness language. In a companion paper, I made the case that linguistic analysis belongs in TAT interpretation (Leffew, 2017), and I use the paper and the analysis together to sketch a path toward continuous, item-level TAT scoring.

Function words as a second channel.

In the Thematic Apperception Test (Morgan & Murray, 1935), a respondent looks at an ambiguous picture and tells a story about it. Under the projective hypothesis, people perceive and organize ambiguous stimuli according to their needs, motives, and feelings (Teglasi, 2010). Clinicians have traditionally read the stories for content, including themes, conflicts, and object relations. Murray held that only expert interpreters could read the test, while McClelland and colleagues developed standardized scoring procedures (Winter, 1998), and researchers continue to question the reliability of picture-story scoring (Gruber & Kreuzpointner, 2013).

Tausczik and Pennebaker (2010) draw a distinction that I carry into projective assessment. Speakers use content words to convey subject matter and function words (pronouns, prepositions, articles, conjunctions, and auxiliary verbs) to connect sentences and to position themselves relative to others, and they choose function words with less deliberate control, which makes those words harder to manipulate. Semin and Fiedler (1988) proposed a finer scheme of four word categories for measuring the abstraction of verbs and adjectives; I use the simpler two-way split. A respondent can avoid mentioning family or money, and the same respondent will find it much harder to control how often they reach for the word I.

With LIWC2015 (Pennebaker, Booth, Boyd, & Francis, 2015), a deterministic dictionary parser, I scored each text on roughly ninety psychometrically anchored categories, including pronoun families, affect, social processes, cognitive processes, drives, and personal concerns. Pennebaker and King (1999) found that LIWC scores and traditional TAT coding correlate on achievement and motivation, and they treat word counting as an addition to human interpretation. I built the pipeline in the same spirit, to give the clinician a second channel of evidence drawn from the parts of language the respondent is not watching.

Six hundred seventy-seven narratives, split four ways.

The narratives come from the Hawkins TAT Project, a collaborative study at Fielding Graduate University in which I led the data preparation and analysis; other project members collected the original protocols. The protocols come from 69 participants and cover 31 card types. I cleaned each narrative to remove assessor prompts and notes on observed behavior, discarded narratives whose card labels I could not verify, and split the text so I could query the same dataset at four grains (each response to each card, each participant’s full protocol, all responses to a single card, and the whole corpus as a single document). I stored LIWC’s published base rates for blogs, expressive writing, novels, and other genres beside the TAT data, so I could check any finding against an external reference.

677 Card responses from 69 participants

31 Cards represented

~90 LIWC2015 categories scored

Card frequencies range from 60 (Card 2) down to 2 (Card 13G), so the card-level analysis stays with the higher-frequency cards. Each row in the dataset carries one of five mutually exclusive types (individual narrative, one card across all respondents, one respondent across all cards, whole corpus, and published base rate), which lets you examine the same LIWC variable at whichever grain a question requires; if you build a dataset like this one, that row-type column is the piece worth adding first.

Card pull by theme.

I ran a MANOVA across Cards 1, 2, 3BM, 4, and 8BM (the five highest-frequency cards, 51 to 60 narratives each) for each of four LIWC families (emotion, social processes, drives, and personal concerns), and all four MANOVAs were significant at p < .001. In the follow-up ANOVAs, 15 of the 18 categories differed by card at p < .05, 10 of them at p < .001; risk, money, and religion did not differ by card. I found that each card’s profile matched what clinicians report the card elicits.

Against the average narrative in the whole corpus, respondents used 1.72 times the achievement language for Card 1 (boy with violin), 1.92 times the family language and 2.52 times the work language for Card 2 (farm scene), 2.74 times the sadness language and 1.51 times the anxiety language for Card 3BM (huddled figure), 2.13 times the anger language for Card 4 (woman restraining a man), and 3.18 times the death language for Card 8BM (surgical scene). I computed both sides of each ratio as means of per-narrative LIWC percentages, so short and long stories count equally on both sides.

Radar chart of six LIWC axes (anxiety, anger, sadness, achievement, power, death) for TAT Cards 1, 3BM, and 8BM, each axis normalized to its maximum across the five compared cards.

Three of the five compared cards on six LIWC axes. Stories about Card 1 scored highest on achievement, stories about Card 3BM on sadness and anxiety, and stories about Card 8BM on death and power. Values are mean LIWC scores per card, each axis normalized to its maximum across Cards 1, 2, 3BM, 4, and 8BM.

Pronouns and negation as clinical signal.

Researchers have mapped function-word patterns onto clinical states. Depressed and suicidal individuals use more first-person singular pronouns, more death-related words, and more negatively valenced language (Bernard, Baddeley, Rodriguez, & Burke, 2016; Tausczik & Pennebaker, 2010); people grieving a romantic loss use more first-person singular pronouns and fewer causal words as their grief deepens (Boals & Klein, 2005); and Hargitai and colleagues (2007) link high rates of negation to the self-aggression thought to accompany suicidality and substance dependence. A clinician has to read these markers in context, since a high rate of first-person singular can mark pain (Tausczik & Pennebaker, 2010), pathological egocentrism (Hargitai et al., 2007), or self-awareness (Seih, Beier, & Pennebaker, 2016).

Because TAT respondents produce a substantial volume of self-generated speech, a clinician can detect markers of psychotic speech in the narratives, including neologisms (Kuperberg, 2010), tangentiality (Marini et al., 2008), and the linking of words that share only temporal proximity, a pattern central to the Cloze procedure (Salzinger, Portnoy, & Feldman, 1964). With LIWC scores beside the narratives, a clinician can better separate disordered thought form from disordered thought content, a distinction that guides both conceptualization and intervention (Langdon, Coltheart, Ward, & Catts, 2002).

For trait measurement, researchers have used linguistic analysis of narratives to distinguish person-oriented from thing-oriented respondents (McIntyre & Graziano, 2016), assessed Big Five traits with latent semantic analysis (Kwantes, Derbentseva, Lam, Vartanian, & Marmurek, 2016), and detected positive and negative affectivity and behavioral inhibition and activation in natural speech (Cohen, Minor, Baillie, & Dahir, 2007). For cognitive functioning, Turk, Brown, Symington, and Paul (2010) used the linguistic content of TAT narratives to distinguish participants with agenesis of the corpus callosum from age- and intelligence-matched controls, a use case beyond the affective domain.

Toward continuous item-level TAT scoring.

The card-level effects above sketch more structure for TAT interpretation; if respondents reliably use a known set of LIWC variables for each card, a clinician can treat those variables as the card’s measurement axes and weigh a respondent’s score on each axis against a card-specific norm. Is this person’s response to Card 3BM unusually sad compared with other people’s responses to Card 3BM? Does the same elevation appear across the full protocol? Does it line up with a current or historical diagnosis?

The frame is closer to continuous manifest-variable item response than to traditional projective scoring; each card becomes its own measure with its own threshold, and the protocol becomes a larger measure built from those item-level scores. The clinician stays in the loop throughout; irony, idiom, and the contextual oddness of a narrative that draws clinical attention through its strangeness lie outside what a word count measures, and Aronow, Weiss, and Reznikoff (2001) place that personalized understanding at the center of TAT interpretation. With the function-word channel, the clinician has a second set of numbers to check a reading against, drawn from the words the respondent is not watching.

Limitations

The analysis is exploratory. The 69 participants are a convenience sample, and the card-level tests cover the five cards with 51 to 60 narratives each; respondents gave fewer than 20 responses to 19 of the other 26 cards, so those cards sit outside the inferential work. The follow-up ANOVAs ran uncorrected across the 18 categories tested, so the effect sizes read best as a map of where card pull is strongest. A word count misses irony, negation in context, and idiom, and word choice varies with education and socioeconomic status (He, Veldkamp, Glas, & de Vries, 2015), so the results describe aggregate language by card; reading an individual respondent’s clinical state from the item-level scoring sketched above needs card-specific norms from a larger, diagnostically characterized sample, and that is the next study.

References

Aronow, E., Weiss, K. A., & Reznikoff, M. (2001). A practical guide to the Thematic Apperception Test: The TAT in clinical practice. Routledge.
Bernard, J. D., Baddeley, J. L., Rodriguez, B. F., & Burke, P. A. (2016). Depression, language, and affect: An examination of the influence of baseline depression and affect induction on language. Journal of Language and Social Psychology, 35(3), 317-326.
Boals, A., & Klein, K. (2005). Word use in emotional narratives about failed romantic relationships and subsequent mental health. Journal of Language and Social Psychology, 24(3), 252-268.
Cohen, A. S., Minor, K. S., Baillie, L. E., & Dahir, A. M. (2007). Clarifying the linguistic signature: Measuring personality from natural speech. Journal of Personality Assessment, 90(6), 559-563.
Gruber, N., & Kreuzpointner, L. (2013). Measuring the reliability of picture story exercises like the TAT. PLoS One, 8(11), e79450.
Hargitai, R., Naszódi, M., Kis, B., Nagy, L., Bóna, A., & László, J. (2007). Linguistic markers of depressive dynamics in self-narratives: Negation and self-reference. Empirical Text and Culture Research, 3, 26-38.
He, Q., Veldkamp, B. P., Glas, C. A., & de Vries, T. (2015). Automated assessment of patients’ self-narratives for posttraumatic stress disorder screening using natural language processing and text mining. Assessment, 1-16. https://doi.org/10.1177/1073191115602551
Kuperberg, G. R. (2010). Language in schizophrenia part 1: An introduction. Language and Linguistics Compass, 4(8), 576-589.
Kwantes, P. J., Derbentseva, N., Lam, Q., Vartanian, O., & Marmurek, H. H. (2016). Assessing the Big Five personality traits with latent semantic analysis. Personality and Individual Differences, 102, 229-233.
Langdon, R., Coltheart, M., Ward, P. B., & Catts, S. V. (2002). Disturbed communication in schizophrenia: The role of poor pragmatics and poor mind-reading. Psychological Medicine, 32, 1273-1284.
Leffew, H. (2017). Interpretation of the Thematic Apperception Test: A case for the inclusion of linguistic analysis [Unpublished manuscript]. Fielding Graduate University.
Marini, A., Spoletini, I., Rubino, I. A., Ciuffa, M., Bria, P., Martinotti, G., … Spalletta, G. (2008). The language of schizophrenia: An analysis of micro and macrolinguistic abilities and their neuropsychological correlates. Schizophrenia Research, 105(1-3), 144-155.
McIntyre, M. M., & Graziano, W. G. (2016). Seeing people, seeing things: Individual differences in selective attention. Personality and Social Psychology Bulletin, 42(9), 1258-1271.
Morgan, C. D., & Murray, H. A. (1935). A method for investigating fantasies: The Thematic Apperception Test. Archives of Neurology & Psychiatry, 34, 289-306.
Pennebaker, J. W., Booth, R. J., Boyd, R. L., & Francis, M. E. (2015). Linguistic Inquiry and Word Count: LIWC2015 [Computer software]. Pennebaker Conglomerates.
Pennebaker, J. W., & King, L. A. (1999). Linguistic styles: Language use as an individual difference. Journal of Personality and Social Psychology, 77(6), 1296-1312.
Salzinger, K., Portnoy, S., & Feldman, R. S. (1964). Verbal behavior of schizophrenic and normal subjects. Annals of the New York Academy of Sciences, 105, 845-860.
Seih, Y. T., Beier, S., & Pennebaker, J. W. (2016). Development and examination of the linguistic category model in a computerized text analysis method. Journal of Language and Social Psychology, 1-13. https://doi.org/10.1177/0261927X16657855
Semin, G. R., & Fiedler, K. (1988). The cognitive functions of linguistic categories in describing persons: Social cognition and language. Journal of Personality and Social Psychology, 54, 558-568.
Tausczik, Y. R., & Pennebaker, J. W. (2010). The psychological meaning of words: LIWC and computerized text analysis methods. Journal of Language and Social Psychology, 29(1), 24-54.
Teglasi, H. (2010). Essentials of TAT and other storytelling assessments (2nd ed.). John Wiley & Sons.
Turk, A. A., Brown, W. S., Symington, M., & Paul, L. K. (2010). Social narratives in agenesis of the corpus callosum: Linguistic analysis of the Thematic Apperception Test. Neuropsychologia, 48(1), 43-50.
Winter, D. G. (1998). “Toward a science of personality psychology”: David McClelland’s development of empirically derived TAT measures. History of Psychology, 1(2), 130-153.