Blog
14/11/2023
Concept mapping is a widely used method for organizing and visualizing the collective understanding of a specific issue among a group of participants (read article: What is concept mapping?). This method, originally developed by Trochim (1989), typically involves three general steps:
- gathering ideas (or concepts) describing the issue at stake
- sorting these ideas into groups according to their general meaning
- identifying clusters representing general themes
For this third step, the ideas are first projected onto a two-dimensional space using non-metric multidimensional scaling (nMDS). Next, a clustering analysis is performed to identify general themes based on the groupings made by the participants. This approach is defined by Trochim in his seminal paper, and it remains the standard in concept mapping today.
However, Péladeau et al. (2017) propose that a more valid approach would be to apply cluster analysis directly to the raw distances calculated from participants' sorting of ideas. Their main argument is that nMDS distorts the data, which in turn influences and weakens the validity of clustering analysis.
To assess the performance of this alternative method, Polygon carried out a study in which we simulated distance matrices with more or less clustered data. We then compared Trochim's method (1989) with the alternative method proposed by Péladeau et al. (2017). Preliminary findings show a better performance of the alternative method, with a better identification of the cluster structure defined a priori, when the partitioning is less precise and the distortion between the original distances and those calculated by nMDS increases1. This is all the more relevant given that empirical concept mapping data display similar properties.
Clustering analysis on the raw distances from the idea sorting offers interesting possibilities to improve concept mapping validity by avoiding the potential distortions introduced by nMDS. This alternative method allows us to better grasp the conceptual relations between ideas, making it a potentially more useful method in research, strategic planning, and decision-making. This is why, at Polygon, we are choosing this approach to improve the validity of our solution.
We aim to publish the results of this study soon in a scientific journal. We are also exploring other ways of improving the method, namely by optimizing the visualization of clusters. Learn more about our other methodological developments in terms of visualizations here (read article: Dimensionality reduction: Alternative approaches for concept mapping). Our methodological developments will also be made available shortly in an open-source Python package (read article: An open source package for concept mapping analysis).
Notes:
1. Calculated with the stress indicator used to identify the best projection in nMDS (https://en.wikipedia.org/wiki/Multidimensional_scaling).
Bibliography:
Péladeau, N., Dagenais, C., & Ridde, V. (2017). Concept mapping internal validity: A case of misconceived mapping? Evaluation and Program Planning, 62, 56–63. https://doi.org/10.1016/j.evalprogplan.2017.02.005
Trochim, W. M. (1989). An introduction to concept mapping for planning and evaluation. Evaluation and Program Planning, 12 (1), 1–16. https://doi.org/10.1016/0149-7189(89)90016-5
Concept Mapping Tool
Research methods
Participatory research
Cluster analysis
Monte-Carlo method
