Quantitative text analysis of language use in rare dementia support groups

Abstract

AbstractBackgroundQuantitative text analysis facilitates study of internal states of speakers in an interspace between self‐report measures and qualitative analyses. The methodology is well‐established in psychological research, including in numerous support group settings.MethodIn this study, Linguistic Inquiry and Word Count (LIWC2015) was employed to characterise conversations in online multi‐component (peers and professionals) support groups for familial caregivers of people living with young onset, atypical or inherited (rare) dementias. Using a corpus of transcripts (129,616 words) from 14 sessions, the differences between trained facilitators’ (n = 2) and participants’ (n = 14) speech were examined, and the impact of agenda on discussion was investigated. Sessions were semi‐structured, addressing independence and identity, or unstructured, without agenda. Statistical methods included comparison of linguistic features with natural speech, Poisson regression, and a clustering algorithm.ResultMeaningful differences between natural speech and the corpus emerged, including in emotional tone (Tone; d = ‐3.2, p<0.001), level of analytic/informal language (Analytic; d = 0.8, p = 0.01), degree of expertise/confidence (Clout; d = 1.0, p = 0.002), and openness/guardedness (Authentic; d = ‐0.9, p = 0.006) (Figure 1). Differences were observed between facilitators and participants in the descriptive (Figure 1) and inferential (Figures 2&3) statistical methods employed. Differences between unstructured and semi‐structured sessions, including between different agendas in the semi‐structured sessions, were observed. Sessions 2 and 4 focused on the participant while sessions 1 and 3 focused on their loved one (Figure 2). The clustering algorithm categorised participants’ conversational turns into three groups: sharing experience (EXP), self‐reflection (SELF), and group processes (GROUP). Contribution‐types were dispersed throughout sessions (Figure 3).ConclusionResults suggest that LIWC2015 can be used to distinguish support group conversations from natural speech, facilitators from participants, and different sessions based on agenda. The methodology could be adapted for use by other peer‐support services wanting insight into their groups using only session transcripts as raw data.