Content analysis of psychological first aid training manuals via topic modelling

Abstract

Background: Psychological First Aid (PFA) is practiced worldwide. This practice in English is guided through a small collection of training manuals. Despite ubiquitous practice and formal training materials, little is known about what topics are covered and in what depth in these influential manuals. As such, we analyzed the topic structure of these training manuals.

Objective: To model the PFA manuals’ topics with the goal of identifying a set of topics with recurrent themes and evaluating the extent to which each manual demonstrated those themes.

Method: This machine learning study employed an unsupervised topic modelling design using Latent Dirichlet Allocation. The variables are (1) the distribution of a word across documents and (2) the distribution of a word across topics. The level of measurement for all variables is continuous. The unit of analysis is words. Preprocessing and data analysis were carried out using the Orange Data Mining Toolbox (Demšar et al., 2013). This programme is a Python GUI.

Results: Results indicated a ten-topic structure to the universe of the English PFA training manuals. These topics were: (1) Refugees, (2) Orientation Activities, (3) Community-Based Applications, (4) PTSD & Other Psychological Issues, (5) Training Materials, (6) Specific Helper Instructions, (7) PFA Scholarship, (8) MHPSS, (9) General Curriculum, and (10) Australian Specific Delivery. The depth of discourse on each topic varied widely between manuals.

Conclusions: The Academics of the PFA topic shows a strong representation of the corpus and suggests current training manuals have stayed true to its evidence-supported practice. The topic of Community-Based Applications strongly represents the corpus and suggests that training models incorporate community-based applications. The scientific foundation and practical implementation of the training guides are essential elements. Limitations and implications were also discussed.