Poster #150 - Meaningful Linguistic Differences In The Adult Attachment Interview – Testing And Validating Linguistic Categorization

• In Event: 3-106 - Poster Session 11
  In Poster Session: PS 11 Section - Parenting & Parent-Child Relationships

Integrative Statement

The words we use can be seen as an indicator of who we are, as words “convey psychological information over and above their literal meaning” (Pennebaker et al., 2003, p.550). Language also provides the data on which trained coders classify attachment representations in the Adult Attachment Interview (AAI). Recently, studies have shown links between attachment representations and differences in the linguistic structure of AAI discourse (e.g., Cassidy, Sherman, & Jones, 2012; Waters et al., 2016). We aimed to replicate and extend these findings by examining the full range of linguistic categories of the LIWC program for three-way, four-way, and Unresolved/Resolved AAI classifications. Furthermore, we examined the potential of automated coding by developing discriminant functions based on the linguistic categories in a test sample and validating these functions in a validation sample.
AAIs were conducted with 307 first-time pregnant women as part of a longitudinal study. The AAIs were transcribed verbatim and coded with the Main and Goldwyn classification system by certified coders. For the linguistic analyses, transcripts were cleaned to hold only the linguistic information provided by the participant and corrected for non-fluencies. Linguistic analyses were performed using Linguistic Inquiry Word Count (LIWC; Pennebaker, Booth, & Francis, 2007) software, which categorizes words into linguistic categories using an internal dictionary (> 93% word categorization). Umbrella categories were excluded, leaving 56 categories for analysis.
AAIs were assigned to a test sample (N = 154) and a validation sample (N = 153) using stratified random sampling. MANOVAs were performed in the test sample to identify word categories that differed between AAI classifications. This method was chosen for its stringent control of multiple testing. All multivariate tests were significant (three-way: F(112, 188) = 2.09, p < .001; four-way: F(168, 279.74) = 1.91, p < .001, Unresolved/Resolved: F(56, 95) = 1.90, p = .003). The analyses indicated 17 word categories differing for three-way classifications, 21 for four-way classifications, and 11 for the Unresolved/resolved distinctions (see Table 1). Next, we used discriminant function analyses to determine whether the linguistic categories could accurately predict AAI classifications. The language-based classifications resulting from the discriminant functions showed convergence with traditional classifications in 72% of the three-way classifications, 69% of the four-way classifications, and 86% of the Unresolved/Resolved classifications. Finally, we cross-validated the discriminant functions in the validation sample. Findings indicated that the language-based classifications converged for 67% (κ = .42), 60% (κ = .35), and 83% (κ = .27), respectively (see Table 2).
This study confirmed that there are meaningful differences in vocabulary during the AAI between participants with different attachment representations. Furthermore, the cross-validation test indicated that the linguistic profiles were substantially similar across the test and validation samples, which is promising for possible future coding purposes. Next steps would be to assess whether individuals also display these linguistic signatures in speech outside of the AAI and whether these linguistic patterns may be related to determinants and outcomes above and beyond attachment classifications.

Authors

• Marije L. Verhage, Vrije Universiteit Amsterdam
  Presenting Author

• Carlo Schuengel, Vrije Universiteit Amsterdam
  Non-Presenting Author

• Mirjam Oosterman, Vrije Universiteit Amsterdam
  Non-Presenting Author